Recent developments in nonlinear analysis

RECENT DEVELOPMENTS IN NONLINEAR ANALYSIS 7641 tp.indd 1 1/28/10 2:23:20 PM This page intentionally left blank RE...

Author: Ammari H. | et al. (eds.)

59 downloads 1044 Views 2MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

RECENT DEVELOPMENTS IN NONLINEAR ANALYSIS

7641 tp.indd 1

1/28/10 2:23:20 PM

This page intentionally left blank

RECENT DEVELOPMENTS IN NONLINEAR ANALYSIS Proceedings of the Conference in Mathematics and Mathematical Physics Fez, Morocco

28 – 30 October 2008

editors

Habib Ammari École Polytechnique, France

Abdelmoujib Benkirane Faculty of Sciences of Fez, Morocco

Abdelfettah Touzani Faculty of Sciences of Fez, Morocco

World Scientific NEW JERSEY

7641 tp.indd 2

•

LONDON

•

SINGAPORE

•

BEIJING

•

SHANGHAI

•

HONG KONG

•

TA I P E I

•

CHENNAI

1/28/10 2:23:20 PM

Published by World Scientific Publishing Co. Pte. Ltd. 5 Toh Tuck Link, Singapore 596224 USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601 UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE

British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library.

RECENT DEVELOPMENTS IN NONLINEAR ANALYSIS Proceedings of the Conference in Mathematics and Mathematical Physics Copyright © 2010 by World Scientific Publishing Co. Pte. Ltd. All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means, electronic or mechanical, including photocopying, recording or any information storage and retrieval system now known or to be invented, without written permission from the Publisher.

For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not required from the publisher.

ISBN-13 978-981-4295-56-7 ISBN-10 981-4295-56-6

Printed in Singapore.

LaiFun - Recent Development in Nonlinear.pmd1

12/28/2009, 1:28 PM

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

v

PREFACE The Conference of Mathematics and Mathematical Physics was held in Fez, Morocco during the period 28–30 October, 2008. It was part of the 5th Congress of the Scientific Research Outlook and Technology Development in the Arab World. The present volume contains the texts of a selection from among the many interesting presentations delivered at the conference. The main objectives of the conference were to promote exchange of the most recent results and emerging ideas and also to merge expertise of researchers in the field not only from the Arab countries but from around the World. The conference covered in a broad and balanced fashion both the theoretical and applied parts of modern nonlinear analysis. Both the Local Organizing and Programme Committees deserve great thanks in creating a well-run and very productive conference, with an exciting programme of keynote lectures and contributed talks. It is a pleasure to thank all of them for their hard work. The support of the Arab Science and Technology Foundation and its president Dr. Abdalla A. Alnajjar are also gratefully acknowledged.

August 6th, 2009 H. Ammari A. Benkirane and A. Touzani

Ecole Polytechnique, France Faculty of Sciences of Fez, Morocco

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

This page intentionally left blank

recent

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

vii

CONTENTS Preface

v

M/v A. Bahri

1

Nonlinear elliptic equations of infinite order A. Benkirane, M. Chrif, and S. El Manouni Some remarks on a sign condition for perturbations of nonlinear problems A. Benkirane, J. Benouna, and M. Rhoudaf

17

30

On the principle eigencurve of a coupled system A. El Khalil

43

Parabolic inclusions with nonlocal conditions A. Boucherif

54

Periodic solutions of nonlinear parabolic equations with measure data and polynomial growth in |∇u| A. El Hachimi and A. Lamrani Alaoui

72

Existence and L∞ -regularity results for some nonlinear elliptic Dirichlet problems A. Youssfi

88

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

viii

Analysis of a new mixed formulation of the obstacle problem B. Seddoug, A. Addou, and A. Lidouh

101

An obstacle problem via a sequence of penalized problems E. Azroul and M. Rhoudaf

123

Numerical analysis of slopes stability under seismic loading in Lebanon F. Hage Chehade and M. Sadek

141

Existence and uniqueness result for a class of nonlinear parabolic equations with L1 data H. Redwane

150

Existence and uniqueness of solutions of some nonlinear equations in Orlicz spaces and weighted Sobolev spaces L. Aharouch, A. Benkirane, J. Bennouna, and A. Touzani

170

Existence of solutions for variational degenerated unilateral problems 181 L. Aharouch, E. Azroul, and M. Rhoudaf Existence and multiplicity results for some p(x)-Laplacian Neumann problems M. Bendahmane, M. Chrif, and S. El Manouni

205

Positive solutions with changing sign energy to nonhomogeneous elliptic problem of fourth order M. Talbi and N. Tsouli

217

A survey on potential theory on Orlicz spaces N. A¨ıssaoui

234

Back on stochastic model for sandpile N. Igbida

266

On radial solutions for Navier boundary eigenvalue problem with p-biharmonic operator S. El Habib and N. Tsouli

278

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

ix

Floquet states of periodically time-dependent harmonic oscillators Y. Achkar, S. Sayouri, and A. L. Marrakchi

290

Renormalized solutions of nonlinear degenerated parabolic problems: Existence and uniqueness Y. Akdim, J. Bennouna, M. Mekkour, and M. Rhoudaf

327

Solvability of quasilinear degenerated elliptic equations with L1 data Y. Akdim

344

Combined fuzzy sliding mode controller for a class of nonlinear systems Y. Alaoui Hafidi, J. Boumhidi, and I. Boumhidi

355

Author Index

365

This page intentionally left blank

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

1

M/v A. Bahri Department of Mathematics Rutgers University, New Brunswick, NJ 08903, USA E-mail: [email protected] Given a three dimensional closed contact manifold (M 3 , α) and a nowhere singular Morse vector-field v in its kernel, we sketch the construction of the space M/v discussed in [4]. We also introduce spaces of immersed curves in M/v and an action functional on these spaces. This is the first step in the completion of a program aimed at computing the homology for contact forms defined in [2] and [7]. Keywords: M/v; Homology for contact forms.

1. Introduction We consider in this paper a three dimensional closed manifold M and a contact form α on M . We assume that there is a nowhere zero vector-field v in kerα, which we also assume to be Morse-Smale. v might have some hyperbolic orbits around which kerα ”does not turn well”, see [1], I.11, [2]. We sketch in what follows a method in order to compute the contact homology that we have defined in [2], [3]. As we have indicated in earlier papers [3], [4], this computation requires the use of the space M/v, a highly pathological, non Hausdorff space. We thus have to devote some time to define such a space, or subsets of this space in a manner that suits our purpose. The idea here has two sides: on one hand, a proper, acceptable definition of the space of orbits mod v, M/v, cannot be given. But a ”hybrid” representation of this space, using partly a section to v and partly periodic orbits can be provided. The next step is then to consider the Z-structure over such a ”section” provided by the contact structure; namely, our ”section” will be defined in a v-invariant subset of M , where α ”turns well” along v. Accordingly, every point x on a v-orbit originating in our ”section” will have infinitely many

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

2

”coincidence points”, see [1], Definition 0.1, p. I.7 ,[2], Definition 9, p. 196 (a ”coincidence point” of [2] is an ”oriented coincidence point” of [1]; the discrepancy is unfortunate, but meaningless); these are points zk such that kerα has rotated kπ from x to zk , k ∈ Z (2kπ for [2]. We will use here the original terminology of [1]). Let ξ be the Reeb vector-field of α. Assume that β = dα(v, .) is, in this subset of M defined by the v-orbits originating at this ”section”, a contact form with the same orientation as α. We define in what follows path spaces adjusted to this Z structure. These path spaces are the natural generalization of the space of immersed curves in S 2 of Maslov index zero. This space of immersed curves appears in a natural way [1], [2], when we study the standard contact structure of S 3 and we take for v a vector-field defining a Hopf fibration in its kernel. Among all the contact forms of this contact structure, there is a special subset which corresponds to the contact forms of this contact structure which are invariant through the antipodal map. These are the ”symmetric” contact forms of this contact structure. They enjoy additional symmetries and the study of their Reeb vector-fields and their periodic orbits is greatly simplified as a consequence. The antipodal map is a special map that generalizes to the most general framework of a contact structure and of a vector-field v, maybe nonsingular, of its kernel. Accordingly, the notion of a symmetric α generalizes, as we will see, albeit under some restrictions. An averaging procedure (section 2) allows to define such a notion. We then sketch the definition in section 3 (the logical order should have been the reverse one) of the space M/v. We essentially show how to define a fundamental domain for an iteration map along v along e.g an attractive orbit of O1 and we show how we can evolve from there and ”travel” using time maps of the one parameter group of v to the hyperbolic orbits and to the repulsive ones, ”filling” sections etc. In the last section, section 4, we sketch the definition of a ”symmetric” functional Js on a space of curves slightly smaller than the space of immersed curves of M/v and we indicate why Js should become very large or tend to ∞ as we tend to the (hyperbolic to the least, and after some adjustments [6]) traces of the periodic orbits of v in M/v. 2. The path spaces, the nearly symmetric α Let us assume that the space M/v has been defined as an ”orbifold section” to v, possibly with boundary. Typically, we would think of the standard

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

3

contact structure of S 3 , of v as being a Morse-Smale perturbation in its kernel of a vector-field defining a Hopf fibration, with an attractive periodic orbit O1 and a repulsive one O2 , see [5], Theorem 1, to find such a vectorfield v in an almost explicit form. M/v then can be taken to be a disk transverse to O1 , with boundary O2 . If we then consider an immersed C 1 -closed curve x(t) in this ”section”, we can lift it above this ”section” along v into a C 1 -curve y(t) so that y(t) ˙ reads as aξ + bv, a positive and y(t) is derived from x(t) by v-transport. y(t) is not unique, neither is it necessarily a closed curve. Rather, given one of the lifts y(t), all other lifts are indexed by an integer k ∈ Z and derived using the map along the v-orbit which assigns to a point x0 the point xk uniquely defined by requiring that β (thus ξ) has completed k half-revolutions between x0 and xk . In order to have y(t) closed, we may have to ask that x(t) be a path, rather than a closed curve, starting at a point x0 and ending at a point x1 , both in the ”section” and both on the same v-orbit. Let T be the map, defined on the ”interior of the section”, which assigns to x0 of this ”section” the next point x00 on the v-orbit through x0 which belongs again to this ”section”. If the end point x1 of the curve x(t) defined above is equal to T m (x0 ) for suitable values of m and if x(1), ˙ the tangent vector to x(t) at x1 , is m equal to DT (x(0)), ˙ then the curves y(t) will be closed curves as we will see. Not only the curve x(t), t ∈ [0, 1], lifts into y(t) and the corresponding family of closed curves above x(t). The curves in this ”section” defined by T i (x(t)), t ∈ [0, 1], i ∈ Z all lift into the same family of curves y(t). So that we find it natural to introduce: Definition 2.1. the space of curves ΛT m (M/v) (M/v is our ”section”) defined as the set of C 1 -curves in M/v running from a point x0 of M/v to the point T m (x0 ). The map T defines a transformation T∗ of this space and we will denote Λ∗T m (M/v) the quotient of this space by T∗ . Our space could be in fact more specific because M/v is typically a stratified space of dimension 2, with boundary one of the attractive or repulsive orbits. Typically, M/v is a disk, with boundary an attractive periodic orbit O1 for example. We can arrange so that T m , restricted to O1 , is the identity map and that ξ rotates exactly mπ in the v-transport along O1 . It is then natural

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

4

to consider the space (M/v)/O1 , the topological quotient of M/v by its subset O1 and therefore to introduce the space Λ∗T m ((M/v)/O1 ) = Λ∗ of C 1 -curves running in (M/v)/O1 from an initial point x0 to T m (x0 ) mod out by the map T ∗ (defined as above, but acting on these new spaces). Embedded into Λ∗ , we find the space of C 1 -immersed curves Imm∗ . The group S 1 acts on these spaces by time translation. Furthermore, above any given curve in Imm∗ , we find a family of closed curves y(t), t ∈ [0, 1], with y˙ = aξ + bv, only that the constant a might change with the curve y in the family. a does not change if the form α is ”symmetric”, that is if, whenever the v-transport, along a v-orbit from x0 to x1 , maps ξ into λξ, then λ = 1. Of course, given a contact structure, a vector-field v in its kernel and a contact form α in this contact structure, we cannot expect α to be ”symmetric”; neither can we assume, in general, the existence of a ”symmetric” α.

The nearly symmetric α However, after averaging α, we can assume that α is nearly ”symmetric”. This averaging procedure goes as follows: considering a point z0 above M/v, we introduce the points zi , i ∈ [−N, N ], N large. zi is defined by the condition that it is the ith -point on the v-orbit such that α is mapped onto 1 α; λi α from z0 to zi . A candidate in order to replace α at z0 is P N 1 i=−N λi

this is formally an almost symmetric contact form in the same contact structure than α. It is, by Lemma 1 of [6] v-convex (that is, denoting θ this form that has v in its kernel, dθ(v, .) is also a contact form with the same orientation than θ) since each of the forms λi α is v-convex (they are pullbacks of α through v-transport maps); but it has the disadvantage to tend to zero as N tends to ∞ on any v-orbit that is asymptotic to an attractive or a repulsive periodic orbit of ! v. N X λi α is also nearly symmetric and does conThe contact form i=−N

verge on any v-orbit tending at ∞ to attractive or repulsive periodic orbits. It is, however, not necessarily v-convex. Near an attractive or a repulsive periodic orbit of v, a model for (α, v) has been provided in [6] p. 47. This model can be slightly modified so that

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

5

i

λi = γ¯ , with 0 < γ¯ < 1. It follows from this model that

N X

i=−N

λi

!

α is

also v-convex (terminology of [5], Lemma 1) near the attractive or repulsive periodic orbits of v (taking (α, v) according to the model). This ”nearly symmetric” form also ”turns well” along v wherever kerα ”turns well” along v. We can then use, as in [6], the second order differential equation along v: (1)[v, [v, ξ]] = −ξ + γ(s)[ξ, v] − γ 0 (s)ds(ξ)v with s(.) denoting the length along v on a given piece of v-orbit with a given origin; this differential equation allows to modify α, ξ, see Lemma 1 of [6], by rescaling the rotation of kerα along v. Starting from the data that we have near each attractive or repulsive periodic orbit, we can evolve along v-orbits and induce a uniform rotation, thereby deriving a nearly symmetric α outside of small tori around the attractive and repulsive periodic orbits of v, along subsets of M where sections to v can be defined; this nearly symmetric αs is now v-convex, this is embedded ! in the rescaling with the use of (1); it is furthermore equal to N X λi α near the repulsive or attractive periodic orbits of v. i=−N

If there are hyperbolic orbits of v, global sections outside of the attractive and repulsive periodic orbits might not be available. We need therefore to remove, in a first step of in our construction, the hyperbolic orbits and their stable and unstable manifolds. The rescaling and the definition of an almost symmetric αs can be completed on the remaining set. In order to extend the definition of our form to the hyperbolic orbits and their stable and unstable manifolds, we follow the construction of [6]; in [6], this construction was carried near a hyperbolic orbit such that kerα did not ”turn well” along it. The general idea would be to extend it to all hyperbolic orbits. This construction needs to be carried out in great detail in the present framework (with the aim of deriving a nearly symmetric form in the vicinity of these orbits, or having the associated functional, see section 4 below, tend to ∞ as the curves come to intersect one of these hyperbolic orbits). Similarly, although the definition of this nearly symmetric contact form is very precise near the attractive and repulsive orbits, the effect of the rescaling, completed above with the use of (1), on the associated variational problem (to be ”defined” in section 4, below) and the behavior of its critical

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

6

points at infinity near the repulsive or attractive periodic orbits of v need to be thoroughly understood.

3. M/v Let us now enter into more details in the construction of ”M/v”. We assume that v has two periodic orbits, O1 which is a attractive and O2 which is repulsive, and a number of periodic orbits which are hyperbolic; but our v is Morse-Smale, with no cycles. For simplicity, let us assume that we have only two hyperbolic periodic orbits O3 and O4 , with O3 dominating O4 , that is the unstable manifold of O3 intersects the stable manifold of O4 and not vice-versa. We want to define the hybrid object M/v. For this, we consider the trace of the stable manifold of O3 on the boundary ∂T1 of a torus T1 transverse to v around O1 . If the eigenvalues of the Poincar´e-return map at O3 are negative, then the trace of Ws (O3 ) on ∂T1 has only one connected component. If the Poincar´e-return map at O3 has positive eigenvalues, then there are exactly two connected components to this trace because the stable manifold of O3 , when deprived of O3 , has two connected components and they do not intersect. Each of these is an embedded differentiable closed curve. By standard arguments, it follows that either both curves are embedded isotopic closed curves which both read homotopically as ma + nb on the two S 1 -generators of the fundamental group of ∂T1 ; or one or both of them are contractible to a point in ∂T1 . Let us assume that we are in this second case: the results which we derive then can be adapted to the first case. Let us think of the intersection of the stable manifold of O4 with ∂T1 . This intersection, though an embedded differentiable curve, is neither closed nor compact. It could also not be connected. We claim that it is made of a finite number of connected components, each of them being an embedded differentiable closed curve whose closure is obtained by addition of one or both connected components of Ws (O3 ) ∩ ∂T1 . This is a fine point which we need to understand. Let us also assume for simplicity that none of the components of the trace of Ws (O3 ) on ∂T1 is homotopic to zero in ∂T1 . Assuming in the sequel that both components of Ws (O3 ) ∩ ∂T1 are not homotopic to zero, they both read ma + nb, m, n prime to each other (they are then homotopic since they do not intersect). The components of Ws (O4 ) ∩ ∂T1 then ”spiral” towards these two isotopic embedded curves.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

7

3.1. cˆ and the fundamental domain We consider a section cˆ to Ws (O3 ) ∩ ∂T1 in ∂T1 . This section is made of two small embedded pieces of curve defined on two intervals, which are transversal to each of the components of Ws (O3 ) ∩ ∂T1 and which are connected by two other embedded pieces of curves in ∂T1 \(Ws (O3 ) ∪ Ws (O4 )) ∩ ∂T1 . We find a closed differentiable embedded curve transverse to both traces of Ws (O3 ) and Ws (O4 ) on ∂T1 . We now consider the Poincar´e-return map f of v from a section σ to v near O1 containing cˆ. σ needs not be transverse to v at O1 , but it should be everywhere else. The choice is very clear if, denoting b the generator transverse to O1 , m in the couple (m, n) defining the homotopy class of Ws (O3 ) ∩ ∂T1 is non-zero. We can take for σ a disk transverse to O1 in T1 . We assume, without loss of generality, that f (ˆ c) is in σ. f (ˆ c) is drawn on the boundary of the solid torus f (T1 ). f is generated by the one parameter group of v, γs and we thus can write f = γs(.) , where s(.) is an appropriate function. We can consider the family of tori γts(.) (T1 ), t ∈ [0, 1]. They define a family of curves in σ which define a fundamental domain ∆. We iterate this fundamental domain ∆ under positive and negative powers of f . The negative iterations end at O1 . The positive iterates go where they should go, but we are going to track a few portions of ∆ under positive iterations. Observe that cˆ intersects each of the components of Ws (O3 ) ∩ ∂T1 at exactly one point. This point, under positive iterations, will get closer and closer to O3 . Adjusting f nearby O3 to become the Poincar´e-return map of O3 at one of its points z, in an appropriate section σ1 , a portion of ∆ defined by two small transversals in cˆ, f (ˆ c) containing the points of Ws (O3 ) in these sets (there are two of them in each of cˆ, f (ˆ c)) and two other ”vertical” pieces of curves connecting these two couples of points(we thereby find a small ”rectangle” in σ) will reach σ1 under iteration from both ”sides”. Indeed, the ”vertical”curves connecting the points of Ws (O3 ) under (adjusted) iteration will reach O3 on two distinct sides, thus z in σ1 from two distinct sides. Just as in [8], in Morse Theory, when considering a nondegenerate critical point, these two ”vertical” transversals then ”spread” under iteration along Wu (O3 ) ∩ σ1 and its iterates. we will denote this set W˜u (O3 )z . It is clear that we have to add it to ∪f n (∆) in order to define M/v. We now have to evolve to O4 from O1 and from O3 . We may assume that the two small transversals in cˆ to Ws (O3 ) contain all of Ws (O4 ) ∩ cˆ and thus that cˆ outside of these small transversals ”spouses” Ws (O4 ) ∩ T1 without intersecting it. Thus, the iterates under f of Ws (O4 ) ∩ cˆ all go to

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

8

σ1 and, from there, should go to O4 . This situation, the lower stage of the tower of domination, offers a new background which we want to discuss now: the first case is when O3 dominates O4 and does not dominate any other hyperbolic orbit. We first elaborate more on this specific situation. We then consider the case when O3 can dominate more than one, typically two-the arguments then generalizehyperbolic periodic orbits. Let us discuss the first case. Let T be the Poincar´e-return map of O4 , defined on a section σ4 to O4 , at a point of O4 . T is generated by the oneparameter group of v and therefore T is homotopic to the identity map, in the set of invertible two dimensional maps. Thus, the differential of T at the origin has a positive determinant. O4 is hyperbolic, thus the differential of T has two real eigenvalues, of the same sign, one larger than one in absolute value, the other one less than one in absolute value also. The differential of T 2 has only positive eigenvalues. Let us consider Ws (O3 ) ∩ σ1 , which is made of one interval, two intervals I + , I − , after removing the fixed point. These two intervals are part of two half-lines L+ , L− which span through the use of the one parameter group of v all of Wu (O3 ) (after the addition of O3 ). We can imagine that σ1 has been extended via the use of the Poincar´ereturn map of O3 and then the time 1; t-map of v so that it reaches near O4 and ”touches” σ2 (σ1 , after iterations, and σ2 have to intersect since O3 dominates O4 ). 3.2. How O1 dominates O3 : A special case Let us assume that σ1 and σ2 can be built as small hyperbolic neighborhoods of pieces of sections to v in Wu (O3 ) and Ws (O4 ) respectively. I + , I − and L+ , L− have been defined for Wu (O3 ), but they can also be defined for O4 . We denote them J + , J − , P + , P − . We use I ± , J ± , L± , P ± to define these sections. The I’s and J’s are used when we are on the ”sides” of Wu (O3 ) and Ws (O4 ) that intersect, taking I and J until a common point x which we may choose to be in σ2 , near O4 . Observe that Wu (O3 ) and Ws (O4 ) intersect in fact at infinitely many points, a subset of which is derived from x through the use of T . We then claim-and this claim is more important for the complete understanding of this specific configuration rather than for the definition of M/v-: Proposition 3.1. Assume that σ1 and σ2 can be built as small hyperbolic neighborhoods of pieces of sections to v in Wu (O3 ) and Ws (O4 ). Then, there

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

9

is another set of points of intersection generated by a y which is not derived from x by iterations. In fact, the intersection points can be viewed as couples of such points (x, y) together with their iterates. Proof. Indeed, considering T 2 , we know that its differential has positive eigenvalues at zero. Thus T 2 maps J + into J + and J − into J − respectively. Without loss of generality, we may assume that it is J + and I + that intersect at x. We consider then Wu (O3 ) ∩ σ2 and more specifically the subset corresponding to I + . Part of cˆ was made of two small pieces of curves transverse to Ws (O3 ) ∩ ∂T1 . we also considered f (ˆ c), thus the image under f of these two small pieces of curves and we connected the ends of each corresponding pair of intervals by ”vertical” lines; this yields two pairs (V1 , V2 ) and (V3 , V4 ) of ”vertical lines” which we iterate using f . The related lines are again denoted Vi . Each Vi is as close as we please to the ”history”, ”under iteration”, of one of the two points of cˆ ∩ (Ws (O3 ) ∩ ∂T1 ). This ”history” defines a set of lines in Ws (O3 ) ∪ Wu (O3 ) ∪ Ws (O4 ) ∪ Wu (O4 ), in fact in the intersection of these sets with the respective sections σ, σj . each Vi neither intersects Ws (O3 ), nor Ws (O4 ). The smaller fundamental domain defined by the two small ”vertical” lines connecting the two intervals and their images under f spread under iteration and ”fill” σ1 (up to the addition of Wu (O3 ) ∩ σ1 ) and, from there, they ”spread” until they ”touch” σ2 just as σ1 did. From there, we use T and we move to σ2 , which we can view to be bounded on each side by portions of the Vi s. Under T 2 , the I + -portion of Wu (O3 ) ∩ σ1 maps into a half-line. Indeed, the image curve does not intersect the Vi s, because, if it did, then some points of Vi would not,under reverse iteration, go to O1 , but would go to O3 . It is therefore entirely contained into σ2 . If the differential of T at the origin has positive eigenvalues, then we can use T in lieu of T 2 . This half-”line” intersects J + , that is the portion of Ws (O4 ) ∩ σ2 , into at least one point, namely x,hence also at its iterates under T 2 . Under iteration, it ”spreads” and its tangent direction becomes pore and more parallel to Ws (O4 ) ∩ σ2 . If the differential of T at zero has negative eigenvalues, then T 2k+1 (x) is in J − rather than J + and all the points of intersection of J + ∩T 2 (I + ) would then read as T 2k (x) if x and only x spans this intersection. However, the orientation of T 2 (I + ) alternates at consecutive intersection points, going from left to right (according to a certain orientation of Ws (O4 ) ∩ σ2 ) at a point and from right to left at the next point. These consecutive points, because x spans the intersection set, are iterates of each other under T 2 .

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

10

This yields a contradiction because the eigenvalues of the differential of T 2 at zero are both positive. The same argument works with T in lieu of T 2 if the differential of T at zero has positive eigenvalues./ We now have tracked our fundamental domain under evolution.we have understood how one of the two sides of each of σ1 and σ2 are ”filled” by the smaller fundamental domain under iteration. The other sides are ”filled” because cˆ intersects as well the trace of the stable manifolds of O3 , O4 on ∂T1 and we can ”drive” M/v, with a proper choice of f to bring our set under iteration to ”fill” the other sides. We add to these iterated sets the two curves Wu (O3 ) ∩ σ1 , Wu (O4 ) ∩ σ2 . The construction of M/v is nearly completed in this easier framework, when O3 dominates only O4 . We still need to understand how this set behaves near O1 , O2 . 3.3. The general case There is a more complicated case, when O3 dominates more than one hyperbolic orbit; e.g O3 dominates O4 , O5 , both hyperbolic orbits. The construction of M/v is then greatly simplified by the following Proposition: Proposition 3.2. Assume N is a two-dimensional surface transverse to v and intersecting e.g. Wu (O3 ) at a point z which is e.g. on the v-orbit of a point of e.g. L+ . Then N ∩ Wu (O3 ) contains a whole half-line which is an image of L+ through the one parameter group of v. This statement holds when N is a surface with boundary; then L+ is replaced by an interval I + . Proof. The intersection of Wu (O3 ) and N is a transverse intersection, which therefore yields a differentiable manifold of dimension 1 transverse to v. Let us consider the connected component of this intersection containing z. It is a one dimensional manifold that has a natural projection π over L+ /J + . π defines a fibration because any two points of π −1 (`), ` given in L+ , can never coalesce: v is transverse to this manifold. It follows that this fibration extends throughout L+ , unless it is limited by the boundary of N ./ 3.4. Outline of the construction of M/v The construction of M/v is derived from the choice of the curve cˆ on ∂T1 and from Proposition 3.2. A careful choice of cˆ allows us to move from the attractive orbit O1 to the other hyperbolic orbits and to the repulsive orbit (there could be more attractive and repulsive orbits; we are only

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

11

describing a simple case here). As we have seen above, near a hyperbolic orbit O3 which is ”directly” dominated by O1 (i.e there is no intermediate hyperbolic orbit), the iterates of a fundamental domain built using cˆ and its image through the Poincar´e return map of O1 will fill a suitable section of this hyperbolic orbit from ”the two sides” (the process is different if the Poincar´e return map at O3 has positive or negative determinant; in the first case, the trace of the stable manifold of O3 on ∂O4 is connected while, in the second case, it has two connected components). The construction of M/v is therefore very clear near O1 and from there to all such O3 s. Then, from a hyperbolic orbit O3 , we may move to another hyperbolic orbit O4 . There, we use Proposition 3.2 which tells us that, because our iterations of our fundamental domain include a point of the stable manifold of O4 , they will contain all the trace of this stable manifold in an appropriate section. In this way, a full ”side” of this section of O4 will be ”filled”. The other side will be ”filled” either through a similar process, that is starting from O3 ; or directly from cˆ because cˆ is chosen appropriately to intersect the trace of the part of the stable manifold of O4 that goes directly to ∂O1 . The process continues in this way (we add to the sections which we encounter the traces of the hyperbolic orbits Oi s in these sections), until we have exhausted all hyperbolic orbits. We are then left with O2 . Our fundamental domain under iterations will come to O2 in a complicated manner, depending also on the linking number of O1 and O2 , of O2 with the other Oi s. This is the part of the construction of M/v that requires further study, until we know precisely what is involved in this construction and this object becomes thereby a straightforward object to use. Despite the fact that our construction of this object is only sketched, we are going to introduce a function on the space of curves Imm∗ defined in section 1 on this space and study its properties. This should lead us to a method for the computation of our homology [2], [3]. 4. The functional The natural functional to use on the space ofR curves ΛT m (M/v) defined 1 in section 2 is the action functional J(x) = 0 αx (x). ˙ This functional is invariant under T if the contact form α is ”symmetric”. However, the contact form built through the averaging procedure of section 2 is only nearly symmetric; it is not symmetric; and the space M/v of section 3 is very well defined only outside of the periodic orbits of v: for example, if we consider

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

12

the case of the standard contact structure of S 3 and if v is a Morse-Smale perturbation of a vector-field defining a Hopf fibration in kerα, O2 , that is the repulsive periodic orbit, is a ”boundary” for M/v. Let us consider in more details the case of the standard contact structure on S 3 , with v having two periodic orbits, one attractive O1 and the other one repulsive O2 . The fundamental observation in this easier framework is that the ”symmetrized” α at points x close to O1 and O2 (how close depends also on N , the number of iterations of T involved in the ”symmetrization” process) reads αx = λ(x)α0x , λ(x) tending to ∞ as x tends to O1 ∪O2 ; that is the ”symmetrized” α has a coefficient tending to infinity on the standard contact form of S 3 . 4.1. Proposition 4.1 Some more is true; namely: Proposition 4.1. v around O1 and O2 may be arranged so that i/ the contact vector field ξ(x) of the ”symmetrized” α tends to zero in norm as x tends to O1 ∪ O2 . In addition, ii/ denoting ψ be the map which assigns to a point x the next coincidence point ([1], [2]) on the positive v-orbit through x, then, after perturbation, the orbits of ξ do not connect x0 and ψ j (x0 ) for x0 in O1 ∪ O2 and for j ∈ Z. We give below the proof of Proposition 4.1. Using the results of [6], properly generalized, we expect Proposition 4.1 to hold for every nowhere zero Morse-Smale v in kerα. The results of [6] are about the behavior of α around hyperbolic periodic orbits of v along which kerα does not turn well. They of course also apply to hyperbolic orbits around which kerα turns well; that is, it should also be possible in such a case to build ”mountains” around these hyperbolic orbits (essentially, i. of Proposition 4.1 should hold around hyperbolic orbits). But, we can then also hope that such ”mountains” can be built around the regions where kerα does not turn well along v. Symmetrizing α outside of small neighborhoods of such regions, we would try to perturb such a symmetric α so that ii) of Proposition 4.1 would hold. Since there is some hope that Proposition 4.1 generalizes, it is useful to prove that this proposition holds in the simpler case of the standard

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

13

contact structure of S 3 and to indicate, pending the complete and rigorous proof of all details, how the computation of the homology, or the existence of periodic orbits for ξ can be derived from this procedure. Proof. Given an integer N , as x moves closer to the e.g attractive orbit O1 of v, the negative iterates of ψ are expanding maps. Assume that kerα rotates twice (v is a perturbation of one of the Hopf-fibrations vectorfields in the kernel of the standard contact form of S 3 ) along O1 , with a uniform rotation and a uniform coefficient of contraction −1 < γ < 0 after a quarter of a turn along O1 , so that the coefficient of contraction after a full turn is γ 4 . This can be achieved after a suitable perturbation of v, kerα near O1 . The negative iterates of α0 , at a point x0 of O1 therefore read as multiples γ −i α0 . Their sum at the order N (from 0 to −N ) therefore reads −N −1 as 1−γ 1−γ −1 α0 . The coefficient in front of α0 tends clearly to ∞ and this fact cannot be destroyed by the contribution of the positive iterates, since ψ is contracting near O1 . This is the basic phenomenon from which, after some additional work estimating the derivatives along ξ, [ξ, v] of λ (the ”symmetrized” α reads λα0 ), i)follows. For ii), we observe that, given e.g y a ξ1 -piece of orbit of a symmetric (over a limit process, outside of O1 ∪ O2 ) contact form α1 connecting two points x0 and ψ j (x0 ) of e.g O1 , if we perturb this symmetric α1 in the vicinity of a point of this ξ1 -piece of orbit into another, symmetric (the symmetry is along ψ and its iterates) form α2 , there will still be, if the intersection problem satisfies the appropriate transversality conditions, in the vicinity of y a ξ2 -piece of orbit connecting two points x1 and x2 of O1 , one close to x0 , the other one to ψ j (x0 ); but, generically, x2 will not be ψ j (x1 ). ii) follows./ 4.2. M ∗ /v and the symmetric αs We now consider the case of a more general nowhere zero vector-field v in the kernel of α, which we assume to be Morse-Smale, having a number of periodic orbits ∪Oi , some elliptic, the other ones hyperbolic. Using the propositions and the results of the previous section, we define the space M/v. We can also define the space: M ∗ /v = M/v\ ∪ Oi . The averaging procedure for α can be completed on M ∗ /v.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

14

If we start from a point of M \ ∪ Oi , the positive and negative iterates under ψ 2 (the transport map along v mapping a point x0 to the next oriented coincidence point (see [1], [2]) on the v-orbit through x0 ) of a given point end up near the attractive and repulsive orbits of v. It follows that ”averaged” limit forms of α, αs are well defined point-wise on M ∗ /v, but might have, as one can easily see, discontinuity points along the stable and unstable manifolds of the hyperbolic periodic orbits of v. If there are no such hyperbolic orbits, as in the case of the standard contact structure of S 3 , with v a small perturbation of a vector-field in kerα defining a Hopf fibration, then a symmetric form αs is well-defined and continuous, differentiable on M \ ∪ Oi . This contact form αs can be used to define a symmetric functional Js , that is a symmetrized version of J (see section 1) on the curves of Imm∗ which do not intersect ∪Oi . We need therefore to understand the behavior of this functional on the curves of this set that are in the immediate vicinity of curves intersecting ∪Oi . Typically, we would want that such curves are ”far” from being critical points of Js and we would in fact want more: namely, we would want that the functional Js tends to ∞ as we approach such curves. 4.3. Js near the hyperbolic orbits For hyperbolic orbits (and this also should solve the discontinuity issues involved by the hyperbolic orbits in the definition of αs ), we have devised a construction in [6]. This construction was carried around hyperbolic orbits having the property that kerα does not rotate well along them. Using large amounts of rotation near the attractive and repulsive orbits, one could build a contact form in the same contact structure such that its associated contact vector-field became tiny near these orbits. This construction can be carried out around the other hyperbolic orbits as well, that is around the orbits along which kerα turns well. It should imply that a functional J can be built on Imm∗ , extending Js . J should be very large or ∞ on the curves of Imm∗ entering and exiting a small neighborhood of a hyperbolic orbit. 4.4. Three additional observations There are three additional observations that are useful: First, this procedure should work around the regions of M where kerα does not rotate well along v. The hope is that a construction similar to the one introduced in [6] for the corresponding hyperbolic orbits can be

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

15

extended to this framework. Second, i) of Proposition 4.1 above implies that the functional J should be very large or ∞ on a curve of Imm∗ that enters a given neighborhood of an attractive or repulsive periodic orbit of v, then intersects this periodic orbit, then exits this given neighborhood (that is Js should tend to ∞ as we approach such a curve). Third, ii) of Proposition 4.1 should also generalize into the statement that there are no curve made of pieces of orbits of the symmetric ξs up to v-jumps between points x and ψ 2 (x) intersecting at least one Oi (repulsive, attractive, or hyperbolic). This is a weaker result than the results foreseen above which say that J is very large or ∞ at curves crossing ∪Oi . But it should be a useful additional result. This provides a very rudimentary version of a scheme in order to compute the homology defined in [3], [4], [7]. But, to the least, one can see here a program; and a glimmer of a reasonable hope that the non-compactness issues can be overcome in Contact Form Geometry. References 1. R. Adams, Sobolev spaces, AC, Press, New York, (1975) 2. L. Aharouch, E. Azroul and M. Rhoudaf Strongly nonlinear variational parabolic problems in Weighted Sobolev spaces to appear in AJMAA. 3. Y. Akdim, E. Azroul and A. Benkirane, Existence Results for Quasilinear Degenerated Equations Via Strong Convergence of Truncations, Revista Matematica Complutense 17, , N.2, (2004) pp 359-379. 4. Y. Akdim, E. Azroul and A. Benkirane, Existence of Solution for quasilinear degenerated Elliptic Unilateral Problems, Ann. Math. Blaise pascal vol 10 (2003) pp 1-20. 5. Y. Akdim, E. Azroul and A. Benkirane, Existence of solution for quasilinear degenerated elliptic equation, Electronic J. Diff. Equ. Vol 2001 , N71, (2001) pp 1-19. 6. H. Brezis, and F.E. Browder, Strongly nonlinear parabolic initial-boundary value problems , Proc. Nat Acad. Sci. U. S. A. 76 (1976). pp. 38-40. 7. J. Berkovits, V. Mustonen, Topological degree for perturbation of linear maximal monotone mappings and applications to a class of parabolic problems,Rend.Mat.Roma,Ser,VII, 12 (1992), pp. 597-621. 8. L. Boccardo, F. Murat, Strongly nonlinear Cauchy problems with gradient dependt lower order nonlinearity, Pitman Research Notes in Mathematics, 208 (1988), pp. 347-364. 9. L. Boccardo, F. Murat, Almost everywhere convergence of the gradients of solutions to elliptic and parabolic equations, Nonlinear analysis, T.M.A., 19 (1992), n 6, pp. 581-597. 10. A. Dallaglio A. Orsina , Non linear parabolic equations with natural growth condition and L1 data. Nolinear Anal., T.M.A., 27 n1 (1996). pp. 59-73.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

16

11. P. Drabek, A. Kufner and L. Mustonen, Pseudo-monotonicity and degenerated or singular elliptic operators, Bull. Austral. Math. Soc. Vol. 58 (1998), 213-221. 12. P. Drabek, A. Kufner and F. Nicolosi, Non linear elliptic equations, singular and degenerated cases, University of West Bohemia, (1996). 13. A. Kufner, Weighted Sobolev Spaces, John Wiley and Sons, (1985). 14. R. Landes, On the existence of weak solutions for quasilinear parabolic initial-boundary value problems, Proc. Roy. Soc. Edinburgh sect. A. 89 (1981), 217-137. 15. R. Landes, V. Mustonen, A strongly nonlinear parabolic initial-boundary value problems, Ark. f. Math. 25. (1987). 16. R. Landes, V. Mustonen, On parabolic initial-boundary value problems with critical growth for the gradient, Ann. Inst. H. Poincar´e11 (2) (1994) 135-158. ˙ sik ˙ sur les probl`emes elliptiques 17. J. Leray, J.L. Lions, Quelques resultats de Viˇ nonlin´eaires par les m´ethodes de Minty-Browder, Bull. Soc. Math. France 93 (1995), 97-107. 18. J.L. Lions, quelques methodes de r´esolution des probl`emes aux limites non lin´eaires, Dunod et Gauthiers-Villars, 1969. 19. A. Porretta Existence results for nonlinear parabilc equations via strong convergence of truncations, Ann. Mat. Pura. Appl. (1999), pp. 143-172. 20. J. M. Rakotoson A Compactness lemma for quasilinear problems: application to parabolic equations J. Funct. Anal. 106 (1992), pp. 358-374. 21. J. Simon Compact sets in the space Lp (0, T, B), Ann. Mat. Pura. Appl. 146 (1987), pp. 65-96. 22. M. Rhoudaf Existence results for Strongly nonlinear degenerated parabolic equations via strong convergence of truncations with L1 -data th`ese de Doctorat de M.Rhoudaf, Univ. Sidi Mohamed Ben Abdella. F`es, Maroc 2006. 23. E. Zeidler, nonlinear functional analysis and its applications, II A and II B, Springer-Verlag (New York-Heidlberg, 1990).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

17

Nonlinear elliptic equations of infinite order A. Benkirane Department of Mathematics, Faculty of Sciences, Atlas-Fes Fes, 3000, Morocco E-mail: [email protected] M. Chrif Department of Mathematics, Faculty of Sciences, Atlas-Fes Fes, 3000, Morocco E-mail: [email protected] S. El Manouni Department of Mathematics, Faculty of Sciences, Al-Imam University P. O. Box 90950, Riyadh, 11623, Saudi Arabia E-mail: [email protected] In this work, generalized Sobolev spaces are considered. Existence of solutions for strongly nonlinear equation of infinite order of the form Au + g(x, u) = f is established. Here A is an operator from a Sobolev space type to its dual and g(x, s) is a lower order term satisfying a sign condition on s. We consider the case where the data f belongs to L1 . Keywords: Strongly nonlinear problem; Anisotropic equations; Infinite order; Existence of solutions.

1. Introduction Let Ω ⊂ IRN , N ≥ 1, be a bounded domain, aα ≥ 0 and pα > 1 are real numbers for all multi-indices α. The purpose of this paper is to study some anisotropic strongly nonlinear elliptic equations of infinite order, with the data f is in L1 (Ω). Let A an operator of infinite order defined by

Au =

∞ X

|α|=0

(−1)|α| Dα aα |Dα u|pα −2 Dα u .

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

18

Along this paper we will deal with the following Dirichlet problem of the form Au + g(x, u) = f

x ∈ Ω.

(1.1)

Here g is a nonlinear term which has to fulfil a sign condition. If A is a Leray–Lions operator, let us mention that in the isotropic case, several studies have been devoted to the investigation of related problems and a lot of papers have appeared (cf. Benkirane,2 Benkirane et al.4 and Br´ezis et al.7 ). In the anisotropic case, it would be interesting to refer the reader to the works Boccardo et al.,5 Feng-Quan,13 also to the recent works Benkirane et al.3 and Chrif et al.,9 where the authors proved the existence of solutions of some anisotropic elliptic equations of higher order. For problems of infinite order, let us point out that in this direction Dubinskii10 proved, under hypothesis (B1 −B4 ) (see Remark 4.1) and certain monotonicity conditions, the existence of solutions for the Dirichlet problem associated with the equation Au = f in some functional Sobolev spaces of infinite order. Our purpose is to prove the same result for strongly nonlinear equations of infinite order of the form (1.1). 2. Preliminaries Let Ω be a bounded domain in IRN . Further aα ≥ 0, pα > 1 are real numbers for all multi-indices α, and k.kpα is the usual norm in the Lebesgue space Lpα (Ω). For a positive integer m, we define the following vector of real numbers p~ = {pα , |α| ≤ m}, and denote p = min{pα , |α| ≤ m}. Now, let us ¯consider the generalized functional Sobolev space W m,~p (Ω) = {u ∈ Lp0 (Ω), Dα u ∈ Lpα (Ω), |α| ≤ m} equipped with the norm kuk =

m X

|α|=0

kDα ukpα .

(2.1)

We define the space W0m,~p(Ω) as the closure of C0∞ (Ω) in W m,~p (Ω) with respect to the norm (2.1). Note that C0∞ (Ω) is dense in W0m,~p (Ω). Both of W0m,~p (Ω) and W0m,~p (Ω) are reflexive, separable Banach spaces if pα > 1 for ~0 all |α| ≤ m (the proof of this is an adaptation from R. Adams1 ). W −m,p (Ω)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

19 α designs its dual where p~0 is the conjugate of p~ i.e. p0α = pαp−1 for all |α| ≤ m. The Sobolev space of infinite order is the functional space defined by ∞ X pα ∞ ∞ α W (aα , pα )(Ω) = u ∈ C (Ω) : ρ(u) = aα kD (u)kpα < ∞ .

|α|=0

C0∞ (Ω)

We denote by the space of all functions with compact support in Ω with continuous derivatives of arbitrary order. Since we shall deal with the Dirichlet problem, we shall use the functional space W0∞ (aα , pα )(Ω) defined by ∞ X pα ∞ ∞ α W0 (aα , pα )(Ω) = u ∈ C0 (Ω) : ρ(u) = aα kD ukpα < ∞ . |α|=0

In contrast with the finite order Sobolev space, the very first question, which arises in the study of the spaces W0∞ (aα , pα )(Ω), is the question of their nontriviality ( or nonemptiness), i.e. the question of the existence of a function u such that ρ(u) < ∞. Definition 2.1. (Dubinskii10 ) The space W0∞ (aα , pα )(Ω) is called nontrivial space if it contains at least one function which not identically equal to zero, i.e. there is a function u ∈ C0∞ (Ω) such that ρ(u) < ∞. It turns out that the answer of this question depends not only on the given parameters aα , pα of the spaces W ∞ (aα , pα )(Ω), but also on the domain Ω. The dual space of W0∞ (aα , pα )(Ω) is defined as follows W −∞ (aα , p0α )(Ω) = {h =

∞ X

0

(−1)|α| Dα hα , ρ (h) =

∞ X

|α|=0

|α|=0

p0

aα khα kpα0α < ∞},

p0α

α (for where hα ∈ L (Ω) and p0α is the conjugate of pα , i.e., p0α = pαp−1 10 11 more details about these spaces, see Dubinskii and Dubinskii ). By the definition, the duality of the space W −∞ (aα , p0α )(Ω) and W0∞ (aα , pα )(Ω) is given by the relation Z ∞ X hα (x) Dα v(x) dx, hh, vi = aα

|α|=0

Ω

which, as it is not difficult to verify, is correct. Let s be a real positive number, we denote by E(s) the integer part of s and denote p = min{pα , |α| ≤ m}. We need the anisotropic Sobolev embedding result. ¯

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

20

Lemma 2.1. Let Ω be a bounded open subset of IR N . If mp < N then W0m,~p (Ω) ⊂ Lq (Ω) ∀q ∈ [p, p∗ [ with p1∗ =

1 p

−

m N.

If mp = N then W0m,~p (Ω) ⊂ Lq (Ω) ∀q ∈ [p, +∞[. If mp > N then W0m,~p (Ω) ⊂ L∞ (Ω) ∩ C k (Ω) where k = E(m − Moreover, the embeddings are compacts.

N p ).

The proof follows immediately from the corresponding embedding theorems in the isotropic case by using the fact that W m,~p (Ω) ⊂ W m,p (Ω). 3. Main result Let Ω ∈ IRN be a bounded domain. We consider the following strongly nonlinear elliptic equation of infinite order of Dirichlet type ∞ X

(−1)|α| Dα aα |Dα u|pα −2 Dα u + g(x, u) = f

in

Ω,

(3.1)

|α|=0

where aα ≥ 0 and pα > 1 are any sequence of real numbers. Let us define the corresponding function Sobolev space of infinite order by ∞ X aα kDα ukppαα < ∞ . W0∞ (aα , pα )(Ω) = u ∈ C0∞ (Ω) : ρ(u) = |α|=0

We assume the following assumptions (A1 ) For all α, aα ≥ 0 and pα > 1, moreover the sequence (pα ) is bounded (A2 ) The space W0∞ (aα , pα )(Ω) is nontrivial. (G1 ) The function g : Ω×IR 7→ IR is of Carath´eodory, that is, it is measurable in x for each fixed u ∈ R and continuous in u for almost all x ∈ Ω, such that, for all δ > 0, sup |g(x, u)| ≤ hδ (x) ∈ L1 (Ω).

|u|<δ

(G2 ) The function g satisfies the ”sign condition”, i.e. g(x, u)u ≥ 0, for a.e. x ∈ Ω and all u ∈ IR. Remark 3.1. In the setting of anisotropic space of finite order and when the operator A is defined by A(u) =

m X

(−1)|α| Dα Aα (x, Dγ u) |γ| ≤ |α|,

|α|=0

the authors have shown the existence of the solution of the problem (1.1), under the following conditions :

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

21 ~0

(A0 ) A : W0m,~p (Ω) 7→ W −m,p (Ω) is a bounded operator, pseudo-monotone and coercive, i.e., lim

kuk→+∞

hAu, ui = +∞, kuk

pα > 1, for all |α| ≤ m. (G0 ) g : Ω × IR 7→ IR satisfies the Carath´eodory conditions, that is, it is measurable in x for each fixed u ∈ R and continuous in u for almost all x ∈ Ω such that sup |g(x, u)| ≤ hs (x),

|u|<s

for a.e. x ∈ Ω, all s > 0 and some function hs ∈ L1 (Ω). We assume also the ”sign condition” g(x, u)u ≥ 0, for a.e. x ∈ Ω and for all u ∈ IR. We present the following result established in Benkirane,3 which will be mainly used in the proof of our result. Theorem 3.1. (Benkirane et al.3 ) Let m ∈ IN ∗ such that mp > N. Sup~0

pose (A0 ) and (G0 ) are satisfied. Then for all f ∈ W −m,p (Ω), there exists u ∈ W0m,~p (Ω) such that ( g(x, u) ∈ L1 (Ω), g(x, u)u ∈ L1 (Ω) R hAu, vi + Ω g(x, u)v dx = hf, vi, ∀v ∈ W0m,~p (Ω). Now we shall prove our main result.

Theorem 3.2. Let us assume the conditions (A1 ), (A2 ), (G1 ) and (G2 ). Then for all f ∈ L1 (Ω), there exists u ∈ W0∞ (aα , pα )(Ω) such that 1 1 g(x, u) ∈ L R (Ω), g(x, u)u ∈ L (Ω) hAu, vi + Ω g(x, u)vdx = hf, vi f or all v ∈ W0∞ (aα , pα )(Ω). Proof. In order to get our result, we will deal with the following steps

1. We prove the existence of approximate solutions un . 2. We establish the a priori estimates. 3. We prove that un converges to an element u ∈ W0∞ (aα , pα )(Ω) and we finally show that u is the solution of our problem.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

22

Step (1): The approximate problem. Let ϕ ∈ C0∞ (IRN ), such that, 0 ≤ ϕ(x) ≤ 1 and ϕ(x) = 1 for x close to 0. Set x fn (x) = ϕ( )Tn f (x), n with the usual truncation Tn given by ( ξ if |ξ| < n Tn ξ = nξ if |ξ| ≥ n. |ξ| It is clear that |fn | ≤ n for a.e x ∈ Ω. Thus, it follows that fn ∈ L∞ (Ω). Using Lebesgue’s dominated convergence theorem, since fn → f a.e. x ∈ Ω and |fn | ≤ |f | ∈ L1 (Ω); we conclude that fn → f strongly in L1 (Ω). Define the operator of order 2n by A2n (u) =

n X

(−1)|α| Dα aα |Dα u|pα −2 Dα u ,

|α|=0

the operator A2n is clearly monotone and satisfies the growth and the coerciveness conditions. Thanks to Theorem 3.1 (Benkirane et al.3 ), there exists at least one solution un ∈ W0n,~p (Ω) of the following problem ( g(x, un ) ∈ L1 (Ω) and g(x, un )un ∈ L1 (Ω) R (Pn ) hA2n (un ), vi + Ω g(x, un )v dx = hfn , vi ∀v ∈ W0n,~p (Ω).

Step (2): A priori estimates. Set v = un and using (A3 ), (G2 ) and the H¨ older inequality, we deduce the estimates n X aα kDαun kppαα ≤ K (3.2) |α|=0

and

Z

Ω

g(x, un )un dx ≤ K

(3.3)

for some constant K = K(f ) > 0. Consequently, we have kun kW n,~p ≤ K. 0

(3.4)

Then via a diagonalization process, there exists a subsequence still, denoted by un , which converges uniformly to an element u ∈ C0∞ (Ω), also for all derivatives there holds D αun → Dαu uniformly in Ω (for more details we refer to Dubinskii10 ). Step (3): Convergence of problem (Pn ).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

23

There exists a solution un of problem (Pn ), n = 1, 2, . . .. Then by passing to the limit, we have

lim hA2n (un ), vi + lim

n→+∞

n→+∞

Z

Ω

g(x, un )v dx = lim hfn , vi, n→+∞

for all v ∈ W0∞ (aα , pα )(Ω). Since fn → f strongly in L1 (Ω), it is clear that lim hfn , vi = hf, vi

n→+∞

for all

v ∈ W0∞ (aα , pα )(Ω).

Now, we shall prove that lim hA2n (un ), vi = hAu, vi,

n→+∞

for all v ∈ W0∞ (aα , pα )(Ω).

Indeed, let n0 be a fix number sufficiently large (n > n0 ) and let v ∈ W0∞ (aα , pα )(Ω). Set hA(u) − A2n (un ), vi = I1 + I2 + I3 , where

I1 =

n0 X

hAα (x, Dγu) − Aα (x, Dγun ), Dαvi

|α|=0

I2 =

∞ X

|α|=n0 +1

I3 = −

n X

hAα (x, Dγu), Dαvi

|α|=n0 +1

hAα (x, Dγun ), Dαvi,

where Aα (x, ξγ ) = aα |ξα |pα −2 ξα , |γ| ≤ |α|. The aim is to prove that I1 , I2 and I3 tend to 0. On the one hand, since Aα (x, ξγ ) is of Carath´eodory type, I1 → 0, and the term I2 is the remainder of a convergent series, hence I2 → 0. On the other hand, for all ε > 0, there holds k(ε) > 0 (see Br´ezis6 p. 56) such that

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

24

n X

|α|=n0 +1

hAα (x, D un ), D vi ≤ γ

n X

α

|α|=n0 +1 n X

≤ c0

|hAα (x, Dγun ), Dαvi| aα

|α|=n0 +1 n X

≤ c0

Z

Ω

|Dαun |pα −1 |Dαv| dx p −1

aα kDαun kpαα

|α|=n0 +1 n X

≤ εc0

kDαvkpα

p

aα kDαun kpαα

|α|=n0 +1

n X

+c0 k(ε)

p

aα kDαvkpαα

|α|=n0 +1

≤ εc0 K + c0 k(ε)

∞ X

|α|=n0 +1

aα kDαvkppαα ,

where K is the constant given in the estimate (3.2). Since the sequence ∞ X aα kDαvkppαα is the remainder of a (pα ) is bounded, this implies that |α|=n0 +1

convergent series, therefore I3 → 0 holds. Hence hA2n (un ), vi → hA(u), vi as n → +∞

for all

v ∈ W0∞ (aα , pα )(Ω).

Now we prove that lim

n→+∞

Z

g(x, un )v dx = Ω

Z

g(x, u)v dx. Ω

Indeed, we have un → u uniformly in Ω, hence g(x, un ) → g(x, u) for a.e. x ∈ Ω. In view of the Fatou lemma and (3.3), we obtain Z Z g(x, u)u dx ≤ lim g(x, un )un dx ≤ K, n→+∞

Ω

Ω

this implies g(x, u)u ∈ L1 (Ω). On the other hand, for all δ > 0 we have |g(x, un )| ≤ sup |g(x, t)| + δ −1 |g(x, un )un | ≤ hδ (x) + δ −1 |g(x, un )un |. |t|<δ

If E is a measurable subset of Ω and ε > 0, we have Z Z |g(x, un )| dx ≤ hδ (x) dx + δ −1 K, E

E

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

25

where K is the constant of (3.3) which is independent of n. For |E| sufficiently small and δ = 2K ε , we obtain Z |g(x, un )| dx ≤ ε. E

Using Vitali’s theorem we get g(x, un ) → g(x, u) in L1 (Ω). Hence it follows that g(x, u) ∈ L1 (Ω). By passing to the limit, we obtain Z hAu, vi + g(x, u)v dx = hf, vi, for all v ∈ W0∞ (aα , pα )(Ω). Ω

Finally, we conclude that 1 1 g(x, u) ∈ L R (Ω), g(x, u)u ∈ L (Ω) hAu, vi + Ω g(x, u)v dx = hf, vi, for all v ∈ W0∞ (aα , pα )(Ω). This completes the proof.

Remark 3.2. Let us point out here that the result in Theorem 3.2, is established with out assuming the regularity of domain, also without any growth restrictions on pα for all multi-indice α. Example 3.1. A prototype example of our problem is defined by ∞ X

(−1)|α| Dα aα |Dα u|pα −2 Dα u + u|u|r h(x) = f

|α|=0

with r > 0, h ∈ L1 (Ω), h(x) ≥ 0 a.e. x ∈ Ω and aα ≥ 0, pα > 1 are real numbers such that the space W0∞ (aα , pα )(Ω) is nontrivial. Since h ∈ L1 (Ω) and h(x) ≥ 0 a.e. x ∈ Ω, the function g(x, u) = u|u|r h(x) satisfies the assumptions (G1 ) and (G2 ). Example 3.2. The following example of an operator of infinite order is closely inspired from the one used in.10 Let us consider the operator Au = [cosD]u(x)

x ∈ IRN , N ≥ 2.

Formally we have Au(x) =

∞ X (−1)n 2n D u(x). 2n! n=0

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

26

The Dirichlet type problem given by ∞ X (−1)n 2n D u + g(x, u) = f x ∈ IRN , 2n! n=0 1 , 2)(Ω), when have a solution in the nontrivial space of infinite order W0∞ ( 2n! 1 f is an element in L (Ω) and g is a function satisfying the assumptions (G1 ) and (G2 ).

4. Concluding remarks Remark 4.1. Let us consider a more general problem, when A is an operator of infinite order defined by A(u) =

∞ X

(−1)|α| Dα Aα (x, Dγ u) |γ| ≤ |α|,

|α|=0

where Aα : Ω × IRλα 7→ IR is a real function, with λα denotes the number of multi-indices γ such that |γ| ≤ |α|, satisfying the following assumptions: (B1 ) Aα (x, ξα ) is a Carath´eodory function for all α, |γ| ≤ |α|. (B2 ) For a.e. x ∈ Ω, all m ∈ IN ∗ , all ξγ , ηα , |γ| ≤ |α| and some constant c0 > 0, we assume that m m X X ≤ c0 aα |ξα |pα −1 |ηα |, A (x, ξ )η α γ α |α|=0

|α|=0

where aα ≥ 0, pα > 1 are reals numbers for all multi-indices α, and for all bounded sequence (pα )α . (B3 ) There exist constants c1 > 0, c2 ≥ 0 such that for all m ∈ IN ∗ , for all ξγ , ξα ; |γ| ≤ |α|, we have m X

|α|=0

Aα (x, ξγ )ξα ≥ c1

m X

|α|=0

aα |ξα |

pα

− c2 .

(B4 ) The space W0∞ (aα , pα )(Ω) is nontrivial. As an open problem, we propose to investigate, under assumptions above (B1 )-(B4 ), the existence result of problem Au + g(x, u) = f

x ∈ Ω,

where g is a nonlinear function and f is an element in L1 (Ω).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

27

Remark 4.2. In the lot of articles, Dubinskii10–12 considered the Sobolev spaces of infinite order corresponding to boundary value problem for differential equations of infinite order and obtained the solvability of these problems in the case where the coefficients of the equation grow polynomially with respect to the derivatives. Dyk Van extends the results of Dubinskii to include the case of operators with rapidly or slowly increasing coefficients (see Dyk Van8 ). In their works, Tran Duk Van et al.16 introduced Sobolev-Orlicz spaces of infinite order and investigated their principal properties. They also established the existence and uniqueness of solutions of some Dirichlet problems for nonlinear differential equations of infinite order. In particular, let Ω a bounded domain in IRN , N ≥ 1, with boundary ∂Ω. Consider the Dirichlet problem defined by ( P |α| α α Au(x) = ∞ x ∈ Ω, |α|=0 (−1) D Aα (x, ..., D u) = f (x) (P b) ω D u(x) = 0, x ∈ ∂Ω, |ω| = 0, 1, .... Here Aα : Ω × IRλα 7→ IR is a real function, λα denotes the number of multi-indices γ such that |γ| ≤ |α|, satisfying the following assumptions: (H1 ) There exist an N −function φα , a function aα ∈ L{φα , Ω}, a continuous bounded function c1α , (1 ≤ c1α (|t|) ≤ const) and a constant b > 0 such that −1

|Aα (x, ξ)| ≤ aα (x) + bφα φα (c1α (|ξα |)ξα ) where ∞ X

|α|=0

kaα kφα < +∞.

(H2 ) There exist functions bm ∈ L1 (Ω), gα ∈ E{φα , Ω}, a continuous bounded function c2α , (c2α (|t|) ≥ c1α (|t|)) and a constant d > 0 such that X X (Aα (x, ξ) − gα (x))ξα ≥ d φα (c2α (|ξα |)ξα ) − bm (x), |α|=m

|α|=m

where ∞ X

|α|=0

kgα kφα < +∞

and

∞ Z X

|α|=0

Ω

|bm (x)| dx < +∞.

(H3 ) The N −functions φα are such that the Sobolev-Orlicz space LW0∞ (φα , Ω) is nontrivial.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

28

(H4 ) For all ξ = (ξ0 , ..., ξα ) and ξ 0 = (ξ00 , ..., ξα0 ) such that ξ 6= ξ 0 we have the inequality m X

|α|=0

(Aα (x, ξ) − Aα (x, ξ 0 ))(ξα − ξα0 ) ≥ 0.

The corresponding functional setting is the Sobolev-Orlicz of infinite order given by LW0∞ (φα , Ω) = {u ∈ C0∞ (Ω) : kuk∞ < +∞}, where kuk∞ = inf{k > 0 :

∞ Z X

|α|=0

Dα u ) dx ≤ 1}. k

φα ( Ω

The dual space of LW0∞ (φα , Ω) is defined by EW

−∞

(φα , Ω) = {f : h(x) =

∞ X

(−1)|α| Dα fα (x)},

|α|=0

where fα ∈ Eφα (Ω)

for all multi-indice α,

and ρ0 (f ) =

∞ X

|α|=0

kfα kφα < +∞.

The duality of the spaces LW0∞ (φα , Ω) and EW −∞ (φα , Ω) is determined by the expression ∞ Z X hf, vi = fα (x)Dα v(x) dx, |α|=0

Ω

which is obviously correct. (For more details about the definition of N −function and Sobolev-Orlicz spaces of infinite order we refer to Tran Duk Van et al.16 ) When the data f belongs to the dual, under assumptions (H1 )-(H4 ) the authors in Tran Duk Van et al.16 proved the existence and uniqueness of the solution of the nonlinear problem (P b). The same approach allows us to deal with boundary value problem (1.1) in the case of Sobolev-Orlicz spaces of infinite order using operators satisfying (H1 ) − (H4 ). We thus obtain the existence result related to that of Theorem 3.2.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

29

Acknowledgments The research of S. El. Manouni is supported by Al-Imam University project No. 28/12. References 1. R. Adams, Sobolev Spaces, Academic, New York, 1975. 2. A. Benkirane, Approximations de type Hedberg dans les espaces W m L log L(Ω) et applications, Ann. Fac. Sci. Toulouse Math. (5) 11 (1990)(2), pp. 67-78. 3. A. Benkirane, M. Chrif and S. El Manouni, Existence results for strongly nonlinear elliptic equations of infinite order, Z. Anal. Anwend. (J. Anal. Appl.) 26 (2007), pp. 303-312. 4. A. Benkirane and J.P. Gossez, An approximation theorem in higher order Orlicz-Sobolev spaces and applications, Studia Math. 92 (1989), pp. 231-255. 5. L. Boccardo, T. Gallouet, and P. Marcellini, Anisotropic equations in L1 , Diff. Int. Equations 9 (1996), pp. 209-212. 6. H. Br´ezis, Analyse Fonctionnelle. Th´eorie et Applications. Paris: Masson 1986. 7. H. Br´ezis and F. E. Browder, Some properties of higher order Sobolev spaces, J. Math. Pures Appl. (9) 61 (1982), pp. 245-259. 8. Chan Dyk Van, Traces of functions from Sobolev-Orlicz classes of infinite order and nonhomogenous value problems for equations with arbitrary nonlinearity, Soviet Math. Dokl. 22 (2) (1980), pp. 626-630. 9. M. Chrif and S. El Manouni, On a strongly anisotropic equation with L1 data, Appl. Anal. 87(7) ( 2008), pp. 865-871. 10. Ju. A. Dubinskii, Sobolev Spaces of Infinite Order and Differential Equations, Teubner-Texte Math. Band 87. Leipzig: Teubner, 1986. 11. Ju. A. Dubinskii, Sobolev spaces for infinite order and the behavior of solutions of some boundary value problems with unbounded increase of the order of the equation, Math. USSR-Sb. 27 (1975)(2), pp. 143-162. 12. Ju. A. Dubinskii, On trace of functions in Sobolev spaces of infinite order and non-homogeneous boundary value problems; Math. USSR Sbornik. 34 (51) (1978), pp. 264-644. 13. L. Feng-Quan, Anisotropic elliptic equations in Lm , J. Convex Anal. 8(2)(2001), pp. 417-422. 14. J. P. Gossez, Some approximation properties in Orlicz Sobolev spaces, Studia Math. 74 (1982), pp. 17-24. 15. J. L. Lions, Quelques M´ethodes de R´esolution des Probl`emes aux Limites Non Lin´eaires, Paris: Dunod; Gauthier-Villars, 1969. 16. Tran Duk Van, R. Gorenflo, Le Van Hap; Sobolev-Orlicz spaces of infinite order and nonlinear differential equations. J. Analysis 10 (1990), pp. 231-245.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

30

Some remarks on a sign condition for perturbations of nonlinear problems A. Benkirane, J. Benouna, and M. Rhoudaf∗ Facult´ e des Sciences Dhar-Mahraz, B.P 1796 Atlas F` es, Morocco E-mail: ∗ [email protected] In this paper, we shall be concerned with the existence result of the quasilinear elliptic equations of the form, Au + g(x, u, ∇u) = f, where A is a Leray-Lions operator from W01,p (Ω) into its dual. On the nonlinear lower order term g(x, u, ∇u), we assume that it is a Carath´eodory function having natural growth with respect to |∇u|, but without assuming the sign condition. The main novelty of our work is a new technique based on a Poincar´e’s 0 inequality. The right hand side f belongs to W −1,p (Ω) . Keywords: Quasilinear elliptic equation; Leray-Lions operator; Existence.

1. Introduction Let Ω be a bounded open subset of IRN , N ≥ 2. Let p be a real number, with 1 < p < +∞, and let p0 be its conjugate H¨ older exponent (i.e. 1p + p10 = 1). Let us consider the following nonlinear elliptic problem Au + g(x, u, ∇u) = f.

(1) 0

Where A is a Leray-Lions operator from W01,p (Ω) into its dual W −1,p (Ω) and g(x, u, ∇u) is a nonlinearity which satisfies the following growth condition |g(x, s, ξ)| ≤ b(|s|)γ(x) + h(s)|ξ|p p

with γ ∈ L p−r (Ω), h ∈ L1 (IR), b(|s|) ≤ β|s|r−1 where 0 ≤ r < p and h, b ≥ 0. More precisely, this paper deals with the existence of solutions to the

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

31

following problem  1,p 1  u  Z ∈ W0 (Ω), g(x, u, ∇u) ∈ LZ(Ω),     g(x, u, ∇u)Tk (u − v) dx  a(x, u, ∇u)Tk (u − v) dx + Z Ω Ω   ≤ f Tk (u − v) dx    Ω   1,p ∀ v ∈ W0 (Ω) ∩ L∞ (Ω)

(2)

0

where f ∈ W −1,p (Ω). Our principal goal in this paper is to prove the existence result for the problem (2) without assuming any sign condition on g. Recently Porreta has proved in5 the existence result for the problem (2) but the result is restricted to b(.) ≡ 1. A different approach (without sign condition) was used in,2 under the assumption g(x, s, ξ) = λs − |ξ|2 with λ > 0. We recall also that the authors used in2 the methods of lower and uppersolutions. For the case of sign condition, many important works have appeared during these last decades. Namely.1 In the literature of the same problems, the sign condition play a crucial role in the proof of main result, to overcame this difficultly we use a new test functions (see lemma below) and a new technique based under inequality of poincar´e. 2. Main results 0

Let A be the nonlinear operator from W01,p (Ω) into its dual W −1,p (Ω) defined as

where a : Ω × IR × IR assumptions: (H1 )

N

Au = −div(a(x, u, ∇u)) is a Carath´eodory function satisfying the following

|a(x, s, ξ)| ≤ [k(x) + |s|p−1 + |ξ|p−1 ]

(3)

[a(x, s, ξ) − a(x, s, η)](ξ − η) > 0. for all ξ 6= η ∈ IR N ,

(4)

a(x, s, ξ)ξ ≥ α|ξ|p

(5)

p0

where k(x) is a positive function in L (Ω) and α is a positive constant. Let g(x, s, ξ) be a Carath´eodory function satisfying the following assumptions: (H2 ) |g(x, s, ξ)| ≤ b(|s|)γ(x) + h(s)|ξ|p

(6)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

32

where b(|s|) ≤ β|s|r−1 where 0 ≤ r < p,

p

γ ∈ L p−r (Ω)

(7)

and h : IR → IR+ with h ∈ L1 (Ω) Lemma 2.1. R |t| Let ϕ(t) = c0 te 0

h(|s|) α

ds

, then ϕ0 (t) −

h(|t|) α |ϕ(t)|

Proof. We can tucking c0 = 1 if t ≥ 0 R |t|

h(|s|) α

ϕ0 (t) = e 0 R |t| =e 0

if t ≤ 0

R |t|

h(|s|) α

ϕ0 (t) = e 0 R |t| =e 0

R |t|

h(|s|)

0 α + t h(|t|) α e h(|t|) ds + α |ϕ(t)|

ds

h(|s|) α

h(|s|)

0 α + t h(|t|) α e h(|t|) ds + α |ϕ(t)|

ds

h(|s|) α

R |t|

≥ c0 . ds

(8)

ds

which implies that

ϕ0 (t) −

R |t| h(|t|) |ϕ(t)| = e 0 α

h(|s|) α

ds

≥ 1.

Theorem 2.1. Assume that the assumption (H1 ) and (H2 ) hold and let f 0 belongs to W −1,p (Ω). Then, there exists a measurable function u solution of the following problem:  1,p 1 u (Ω)  Z ∈ W0 (Ω), g(x, u, ∇u) ∈ L Z    g(x, u, ∇u)Tk (u − ϕ) dx a(x, u, ∇u)∇Tk (u − ϕ) dx + (P ) Ω ΩZ    ≤ f Tk (u − ϕ) dx, ∀ϕ ∈ W01,p (Ω) ∩ L∞ (Ω) ∀k > 0. Ω

Step (2) Approximate problem Let us consider the sequence of approximate problem  1,p  u  Zn ∈ W0 (Ω) Z     gn (x, un , ∇un )(un − v) dx  a(x, un , ∇un )∇(un − v) dx + Ω Z Ω (Pn )   fn (un − v) dx ≤      ∀ v ∈ W 1,p (Ω) ∩ΩL∞ (Ω) 0

where

gn (x, s, ξ) =

g(x, s, ξ) . 1 + n1 |g(x, s, ξ)|

(9)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

33

Note that |gn (x, s, ξ)| ≤ |g(x, s, ξ)| et |gn (x, s, ξ)| ≤ n, then for fixed n ∈ IN , the approximate problem (Pn ) has at least one solution.4 Lemma 2.2. Let un be a solution of the problem (Pn ), then we have Z

|∇un |

Ω

p

p1

≤ c,

(10)

where c is a positive constant not depending on n. R |t|

h(|s|)

ds α Proof. Let v = un − ϕ(un ) where ϕ(t) = te 0 (the function h ap1,p pears in (H2 )). Since v ∈ W0 (Ω)) v is admissible test function in (Pn ), then

Z

Ω

a(x, un , ∇un )∇(ϕ(un )) dx +

Z

Ω

gn (x, un , ∇un )ϕ(un ) dx ≤

Z

f ϕ(un ) dx Ω

which implies that, Z Z 0 a(x, un , ∇un )∇un ϕ (un ) dx ≤ b(|un |)γ(x)|ϕ(un )| dx Z Z Ω Ω |f ||ϕ(un )| dx + h(|un |)|ϕ(un )||∇un |p dx + Ω

Ω

0

since ϕ ≥ 0, then Z Z Z h(|un |) |∇un |p (ϕ0 (un ) − |f ||un | |un |r γ(x) + c0 |ϕ(|un |)|) ≤ βc0 α Ω Ω Ω R +∞

with c0 = e 0 we deduce that Z

Ω

h(|s|) α

ds

p

and by using poincar´e inequality and Lemma 2.1,

|∇un | ≤ c

Z

Ω

|∇un |

p

pr

+c

Z

Ω

|∇un |

p

p1

.

(11)

Consequently, since r < p we have Z

Ω

|∇un |p

p1

≤c

(12)

then, we can extracts a subsequence still denote by un such that, un * u weakly in W01,p (Ω)

(13)

un → u strongly in Lp (Ω)

(14)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

34

this yields, by (3), the existence of a function h ∈

a(x, un , ∇un ) * h weakly in

N Y

0

Lp (Ω) such that

i=1

N Y

0

Lp (Ω).

(15)

i=1

Lemma 2.3. Let un be a solution of the problem (Pn ), then we have the following assertions: Assertion (i)

lim lim

j→∞ n→∞

Z

{j≤|un |≤j+1}

a(x, un , ∇un )∇un dx = 0.

(16)

Assertion (ii)

lim lim

j→∞ n→∞

Z

Ω

(a(x, Tk (un ), ∇Tk (un )) − a(x, Tk (u), ∇Tk (u)))

(17)

∇(Tk (un ) − Tk (u))hj (un ) = 0.

Assertion (iii)

Tk (un ) → Tk (u) strongly in W01,p (Ω) as n → ∞.

(18)

Proof assertion (i). Consider the following function v = un − R |un | h(|s|) ds α e0 T1 (un − Tj (un ))+ , for j > 1, then we obtain, Z

h(|un |) R0|un | h(s) ds α e a(x, un , ∇un )∇un T1 (un − Tj (un ))+ α ΩZ R |un | h(s) ds α + a(x, un , ∇un )∇T1 (un − Tj (un ))+ e 0 ZΩ R |un | h(s) ds α + gn (x, un , ∇un )e 0 T1 (un − Tj (un ))+ ZΩ R |un | h(s) ds α ≤ fe 0 T1 (un − Tj (un ))+ . Ω

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

35

From the growth condition (H2 ), we have Z h(|un |) R0|un | h(s) ds α a(x, un , ∇un )∇un T1 (un − Tj (un ))+ e α Ω Z R |un | h(s) ds α a(x, un , ∇un )∇T1 (un − Tj (un ))+ e 0 + Z Ω R |un | h(s) ds α |γ(x)|b(|un |)e 0 T1 (un − Tj (un ))+ ≤ Z Ω R |un | h(s) ds α + h(|un |)|∇un |p e 0 T1 (un − Tj (un ))+ ZΩ R |un | h(s) ds α T1 (un − Tj (un ))+ |f |e 0 + Ω

R |un |

R +∞

h(s)

h(s)

ds ds α α = c0 , gives: which thanks to (5) and 1 ≤ e 0 ≤e 0 Z R |un | h(s) ds α a(x, un , ∇un )∇T1 (un − Tj (un ))+ e 0 Ω Z Z |f |T1 (un − Tj (un ))+ |γ(x)|b(|un |)T1 (un − Tj (un ))+ + c0 ≤ c0 Ω

Ω

which implies that, Z Z a(x, un , ∇un )∇un ≤ c0 β {j≤un ≤j+1}

{|un |>j} Z

+c0

Ω

(19)

|un |r−1 |γ(x)|T1 (un − Tj (un ))+ |f |T1 (un − Tj (un ))+ (20)

we deduce that Z Z a(x, un , ∇un )∇un ≤ c0 β |un |r−1 |γ(x)|T1 (un − Tj (un ))+ {j≤un ≤j+1} Ω Z + |f |T1 (un − Tj (un ))+ Ω p−r Z pr Z p p + p−r p (γ(x)|T1 (un − Tj (un )) |) ≤ c0 β |un | Ω Z Ω + c0 |f |T1 (un − Tj (un ))+ Ω

(21)

by using the inequality poincar´e and (10) we get, Z

{j≤un ≤j+1}

a(x, un , ∇un )∇un ≤ c

Z

+

p p−r

(γ(x)|T1 (un − Tj (un )) |) Z c0 |f |T1 (un − Tj (un ))+ +

Ω

Ω

p−r p (22)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

36

then, by Lebesgue’s theorem the right hand side goes to zero as n and j tends to infinity. Therefore, passing to the limit first in n, then in j, we obtain from (22) Z lim lim a(x, un , ∇un )∇un dx = 0 (23) j→∞ n→∞

{j≤un ≤j+1}

R |un |

h(s)

ds α on the other hand, consider the test function v = un +e− 0 T1 (un − − Tj (un )) in (Pn ), then Z R |un | h(s) ds α T1 (un − Tj (un ))− dx a(x, un , ∇un )∇ −e− 0 ΩZ R |un | h(s) ds α + gn (x, un , ∇un ) −e− 0 T1 (un − Tj (un ))− dx ZΩ R |un | h(s) ds α T1 (un − Tj (un ))− dx ≤ f −e− 0 Ω

which implies according to (H2 ) that, Z h(|un |) − R0|un | h(s) ds α a(x, un , ∇un )∇un T1 (un − Tj (un ))− dx e α ΩZ R |un | h(s) ds α − a(x, un , ∇un )e− 0 ∇T1 (un − Tj (un ))− dx ZΩ R |un | h(s) ds α ≤ b(|un |)e− 0 T1 (un − Tj (un ))− γ(x) dx ZΩ R |un | h(s) ds α + h(|un |)e− 0 T1 (un − Tj (un ))− |∇un |p dx ZΩ R |un | h(s) ds α + |f |e− 0 T1 (un − Tj (un ))− dx. Ω

R |un |

h(s)

ds α ≤ 1 it is possible to conclude that, From (5) and since 0 ≤ e− 0 Z R |un | h(s) ds α − a(x, un , ∇un )e− 0 ∇T1 (un − Tj (un ))− dx Ω Z Z − ≤ b(|un |)T1 (un − Tj (un )) γ(x) dx + |f |T1 (un − Tj (un ))− dx Ω

Ω

then, −

Z

{−j−1≤un ≤−j} Z

≤

− a(x, un , ∇un )∇u− ne

{uZ n ≤−j}

+

Ω

R |un | 0

h(s) α

ds

dx

β|un |p−1 T1 (un − Tj (un ))− γ(x) dx

|f |T1 (un − Tj (un ))− dx

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

37

Consequently, we have Z

{−j−1≤un ≤−j}

≤β

Z

a(x, un , ∇un )∇un

{un ≤−j}

|un |p

! pr

Z

{unZ ≤−j}

+

Ω

(γ(x)T1 (un − Tj (un ))− )

p p−r

! p−r p

|f |T1 (un − Tj (un ))− dx

(24) since (un )n∈IN is bounded in Lp (Ω), we deduce by Lebesgue’s theorem that the terms of the right-hand side of the last inequality goes to zero as n and j tends to infinity. Then, (24) becomes lim lim

j→∞ n→∞

Z

{−j−1≤un ≤−j}

a(x, un , ∇un )∇un dx = 0.

(25)

Finally, combining (23) and (25), we have lim lim

j→∞ n→∞

Z

{j≤|un |≤j+1}

a(x, un , ∇un )∇un dx = 0.

(26)

Proof of assertion (ii). We will use the following function of one real variable, which is defined as follow:  if  1  0 if hj (s) =  j + 1 − s if   s + j + 1 if

Let v = un − e we obtain Z

R |un | 0

h(s) α

ds

|s| ≤ j |s| ≥ j + 1 j ≤s≤j +1 − j − 1 ≤ s ≤ −j.

(Tk (un ) − Tk (u))+ hj (un ) as test function in (Pn ), R |un |

h(s)

ds α a(x, un , ∇un )∇(e 0 (Tk (un ) − Tk (u))+ hj (un )) dx ΩZ R |un | h(s) ds α + gn (x, un , ∇un )e 0 (Tk (un ) − Tk (u))+ hj (un ) dx ZΩ R |un | h(s) ds α ≤ fe 0 (Tk (un ) − Tk (u))+ hj (un ) dx Ω

(27)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

38

using (5) and (H2 ), we obtain Z R |un | h(s) ds α a(x, un , ∇un )∇(Tk (un ) − Tk (u))+ e 0 hj (un ) dx Ω Z R |un | h(s) ds α ≤ b(|un |)γ(x)e 0 (Tk (un ) − Tk (u))+ hj (un ) dx ZΩ R |un | h(s) ds α (Tk (un ) − Tk (u))+ dx + a(x, un , ∇un )∇un e 0 Z{j≤|unR|≤j+1} |un | h(s) ds α (Tk (un ) − Tk (u))+ hj (un ) dx |f |e 0 + Ω

R |un |

1

h(s) α

ds

R +∞

h(s) α

ds

(28) = c0 < +∞, which

since h ∈ L (Ω), we have e ≤e implies that, Z R |un | h(s) ds α a(x, un , ∇un )∇(Tk (un ) − Tk (u))+ e 0 hj (un ) dx Z Ω ≤ c0 β |un |r−1 γ(x)(Tk (un ) − Tk (u))+ hj (un ) dx Z Ω Z +c2 a(x, un , ∇un )∇un dx + c1 |f |(Tk (un ) 0

0

{j≤|un |≤j+1}

Ω

−Tk (u))+ hj (un ) dx

By the Holder’s inequality, we have, Z R |un | h(s) ds α hj (un ) dx a(x, un , ∇un )∇(Tk (un ) − Tk (u))+ e 0 Ω p−r Z rp Z p p p + p−r dx ≤c |un | dx (γ(x)(Tk (un ) − Tk (u)) hj (un )) Ω Z Ω a(x, un , ∇un )∇un dx +c2 Z{j≤|un |≤j+1} |f |(Tk (un ) − Tk (u))+ hj (un ) dx +c1 Ω

applying again (16) and Lebesgue’s theorem the terms of the right hand side of last inequality goes to zero as n and j tend to infinity, then, Z R |un | h(s) ds α lim lim a(x, un , ∇un )∇(Tk (un ) − Tk (u))+ e 0 hj (un ) dx = 0. j→∞ n→∞

Ω

(29)

Moreover, (29) becomes, Z a(x, Tk (un ), ∇Tk (un ))∇(Tk (un ) − Tk (u)) {Tk (u k (u)≥0} Rn|u)−T n | h(s) ds 0 α ×e hj (un ) Z

−

dx

{Tk (un )−Tk (u)≥0,|un |≥k}

a(x, un , ∇un )∇Tk (u)e

R |un | 0

h(s) α

ds

hj (un ) dx

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

39

which gives lim lim

Z

a(x, Tk (un ), ∇Tk (un ))(∇Tk (un ) − ∇Tk (u))

j→∞ n→∞ {T (u )−T (u)≥0} n k k R |un | h(s) ds α ×e 0 hj (un ) dx

Since e

R |un | 0

lim lim

j→∞ n→∞

Z

h(s) α

ds

= 0.

≥ 1, the consequently we can write ,

{Tk (un )−Tk (u)≥0}

[a(x, Tk (un ), ∇Tk (un )) − a(x, Tk (un ), ∇Tk (u))] ×[∇Tk (un ) − ∇Tk (u)]hj (un ) dx = 0

(30)

on the other hand, taking v = un + e−

R |un | 0

h(s) α

ds

(Tk (un ) − Tk (u))− hj (un )

as test function in (Pn ) and reasoning as in (30) it is possible to conclude that, Z lim lim [a(x, Tk (un ), ∇Tk (un )) − a(x, Tk (un ), ∇Tk (u))] j→∞ n→∞

{Tk (un )−Tk (u)≤0}

×[∇Tk (un ) − ∇Tk (u)]hj (un ) dx = 0

(31) Combining (30) and (31), we deduce (17). Proof of assertion (iii). First we have Z [a(x, Tk (un ), ∇Tk (un )) − a(x, Tk (un ), ∇Tk (u))][∇Tk (un ) − ∇Tk (u)] dx ΩZ = [a(x, Tk (un ), ∇Tk (un )) − a(x, Tk (un ), ∇Tk (u))] Ω

Z ×[∇Tk (un ) − ∇Tk (u)]hj (un ) dx + [a(x, Tk (un ), ∇Tk (un )) − a(x, Tk (un ), ∇Tk (u))] Ω

×[∇Tk (un ) − ∇Tk (u)](1 − hj (un )) dx

thanks to (17) the first integral of the right hand side converges to zero as n and j tend to infinity, for the second term, we have for j large enough (j > k) Z [a(x, Tk (un ), ∇Tk (un )) − a(x, Tk (un ), ∇Tk (u))] Ω

Z ×[∇Tk (un ) − ∇Tk (u)](1 − hj (un )) dx = a(x, Tk (un ), ∇Tk (u))∇Tk (u)(1 − hj (un )) dx Ω

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

40

this integral converges to zero since a(x, Tk (un ), ∇Tk (u)) converges to 0 a(x, Tk (u), ∇Tk (u)) strongly in (Lp (Ω))N while ∇Tk (u)(1 − hj (un )) converges to zero strongly in (Lp (Ω))N . Finally, we conclude that Z lim [a(x, Tk (un ), ∇Tk (un ))−a(x, Tk (un ), ∇Tk (u))][∇Tk (un )−∇Tk (u)] = 0 n→∞

Ω

then Lemma 5 of,3 implies that, Tk (un ) → Tk (u) strongly in W01,p (Ω).

(32)

And ∇un → ∇u

a. e. in Ω.

Lemma 2.4. Let be un a solution of (Pn ), then gn (x, un , ∇un ) → g(x, u, ∇u) strongly in L1 (Ω) Proof. Let v = un + e− in (Pn ), then Z

R |un | 0

h(|s|) α

−

ds

R |un |

Z

(33)

0 un

h(|s|) α

h(|s|)χ{s<−h} ds as test function

ds

Z

0

h(|s|)χ{s<−h} ds a(x, un , ∇un )∇ −e un Z Z 0 R |un | h(|s|) ds α h(|s|)χ{s<−h} ds − gn (x, un , ∇un )e− 0 Ω un Z Z 0 R |un | h(|s|) ds α ≤ |f |e− 0 h(|s|)χ{s<−h} ds Ω

Ω

0

un

which implies that, Z Z h(|un |) − R0|un | h(|s|) ds 0 α h(|s|)χ{s<−h} ds e a(x, un , ∇un )∇un α un Ω Z R |un | h(|s|) ds α + a(x, un , ∇un )∇un e− 0 h(|un |)χ{un <−h} dx Ω Z Z 0 R |un | h(|s|) ds α ≤ b(|un |)|γ(x)|e− 0 h(|s|)χ{s<−h} ds Ω un Z 0 Z R |un | h(|s|) ds α h(|s|)χ{s<−h} ds + h(|un |)|∇un |p e− 0 un Z ZΩ 0 R |un | h(|s|) ds α h(|s|)χ{s<−h} ds + |f |e− 0 Ω

un

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

41

using (5) and since Z

Z

0 un

h(|s|)χ{s<−h} ds ≤ R |un |

Z

−h

h(|s|) ds, we get

−∞

h(|s|)

ds α a(x, un , ∇un )∇un e− 0 h(|un |)χ{un <−h} dx Ω Z −h Z −h Z Z |f | h(|s|) ds dx |un |r−1 γ(x) h(|s|) ds + c ≤c Ω −∞ Z−∞ ZΩ−h ≤c h(|s|) ds |un |r−1 γ(x) dx −∞ Z |un |≥1 Z |f | dx |un |r−1 γ(x) dx + c + Ω Z −h |un |≤1 r h(|s|) ds (k∇un k1,p + c0 + kf k1,p0 ) ≤c −∞ Z −h h(|s|) ds ≤c −∞

using again (5), we obtain Z {un <−h}

h(|un |)|∇un |p ≤ c

Z

−h

h(|s|) ds

−∞

and since h ∈ L1 (IR), we deduce that, Z lim sup h(|un |)|∇un |p dx = 0 h→+∞ n∈IN

consider v = un − e

R |un | 0

(34)

{un <−h}

h(|s|) α

ds

Z

un

h(|s|)χ{s<−h} ds as test function in

0

(Pn ), then similarly to (34), we deduce that Z h(|un |)|∇un |p dx = 0 lim sup

(35)

assuming (34) and (35) we obtain Z lim sup

(36)

h→+∞ n∈IN

h→+∞ n∈IN

{un >h}

{|un |>h}

On the other hand, we have Z Z b(|un |)|γ(x)| dx ≤ {un >h}

h(|un |)|∇un |p dx = 0.

{un >h}

≤

Z

≤c

Ω

Z

|un |r−1 |γ(x)| dx

|un |p

rp

{un >h}

Z

{un >h}

|γ(x)|

p

! p1

|γ(x)|

p p−r

! p−r p

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

42

then, we conclude that lim sup

h→+∞ n∈IN

Z

Ω

b(|un |)|γ(x)| dx = 0

(37)

Combining (36), (37) and Vitali’s theorem, we conclude (33) Proof of the theorem. Let ϕ ∈ W01,p (Ω) ∩ L∞ (Ω) and take v = un + Tk (un − ϕ) as test function in (Pn ), we get Z Z a(x, un , ∇un )∇Tk (un − ϕ) dx + gn (x, un , ∇un )Tk (un − ϕ) dx Z Ω Ω (38) ≤ f Tk (un − ϕ) dx. Ω

Finally, from (18) and (33), we can pass to the limit in (38), this completes the proof of the theorem. References 1. A. Bensoussan, L. Boccardo and F. Murat On a non linear partial differential equation having natural growth terms and unbounded solution, Ann. Inst. Henri Poincar´e 5 N4 (1988) 149-169. 2. L. Boccardo, F. Murat and J.P. Puel, L∞ estimate for some nonlinear elliptic partial differential equations and application to an existence result, SIAM J. Math. Anal. 2 (1992) 326-333. 3. L. Boccardo, F. Murat and J.P. Puel, Existence of bounded solutions for nonlinear elliptic unilateral problems, Annali. Mat. Pura Appli. (1988) 183196. 4. J.L. Lions, Quelques m´ethodes de r´esolution des probl`emes aux limites non lin´eaires, Dunod, Paris (1969). 5. A. Porretta, Nonlinear equations with natural growth terms and measure data, Electronic J. Diff. Equ., Conference 09 (2002) 181-202.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

43

On the principle eigencurve of a coupled system A. El Khalil Department of Mathematics, College of Sciences Al-Imam Muhammad Ibn Saud Islamic University P.O. Box 90950, Riyadh 11623, Saudi Arabia E-mail: [email protected] We prove that for any λ ∈ IR, there is a sequence of eigencurves (µk (λ))k for the nonlinear coupled elliptic system   −∆p u = λa(x)|u|α−1 |v|β+1 u + µ|u|α−1 |v|β+1 u in Ω −∆q v = λa(x)|u|α+1 |v|β−1 v + µ|u|α+1 |v|β−1 v in Ω  (u, v) ∈ W01,p (Ω) × W01,q (Ω),

by using min-max methods. We prove also via an homogeneity type condition that the eigenvector corresponding to the principal µ1 (λ) is bounded, positive and smooth. We end this work by proving that µ1 (λ) is simple. Keywords: Quasilinear system; p-Laplacian operator; Principal eigencurve.

1. Introduction Let Ω be a bounded domain in IRN not necessary regular; N > 1, p > 1, q > 1, and α , β > 0 satisfying the homogeneity assumption: α+1 β+1 + = 1, p q

(H)

and a(x) ∈ L∞ (Ω)\{0} be an indefinite weight function which can change the sign. Here, λ and µ are tow parameter eigenvalues. We consider the following nonlinear elliptic system  α−1 |v|β+1 u + µ|u|α−1 |v|β+1 u in Ω  −∆p u = λa(x)|u| α+1 Sp,q (λ) −∆q v = λa(x)|u| |v|β−1 v + µ|u|α+1 |v|β−1 v in Ω  1,p (u, v) ∈ W0 (Ω) × W01,q (Ω).

Problems where the operator −∆p is present arise both from pure and applied mathematics. Like in the theory of quasiregular and quasiconformal mapping, as well as from a variety of applications, e.g. Non-Newtonian

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

44

fluids, reaction diffusion problems, flow through porous media. Nonlinear elasticity, glaciology, petroleum extraction, astronomy,. . . etc. Many frameworks have been dealing with the corresponding equation −∆p u = λ|u|γ−1 u

´lin et al.,7 Th´ one can cite for example Anane,1 Binding,2 The elin;6 and in the particular case p = γ there are many works, we cite El Khalil et al.,9 Fleckinger et al.,10 Hess et al.,11 Kato,12 Lindqvist,13 Richardson,14 . . . etc. elin6 ) where In the system case, we find the work of De Th´elin (cf. Th´ λ = 0 with Ω is regular bounded domain; also El Khalil, Ouanan and Touzani studied the stability with respect to the rheological exponent p and q of the first eigenvalue see [11], Chabrowski Chabrowski3 considers particular case with λ = 0 introducing two perturbation functions in the right hand side of the system; and Flekinger and coauthors studied this problem in Fleckinger et al.10 with µ is some function on x satisfying some large hypothesis with Ω = IRN . In our work, we investigate the situation improving the existence at least a sequence of the eigencurves (eigenpair (λ, µ(λ)) for any bounded domain by using the Ljusternik-Schnirelman theory on C 1 -manifolds cf. Szulkin.16 So we give a new characterization of the principal eigencurve and also we find some result about the associated eigenvector when λ is in a suitable range of IR depending of λ1 = µ1 (0) and the weight function a(x). The rest of this paper is organized as follows. In section 2 we introduce some definitions and prove some technical preliminary, in section 3 we prove that the principal eigencurve is well defined and there exist at least a sequence of the eigencurve, finally we prove some result about the eigenvector associated of µ1 (λ) when λ is in a suitable range. 2. Preliminaries 2.1. Functional framework In the sequel, we shall use the standard notations. W01,p (Ω) is the completion of C0∞ (Ω) in the space W 1,p (Ω). In the other words, W01,p (Ω) = u ∈ W 1,p (Ω) | u|∂Ω = 0 the value of u on Ω beingunderstood in the trace sense. 1,p It is well known that W0 (Ω), k∇.kp is separable, reflexive and uniformly convex, for 0 < p < +∞.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

45 0

1 p

The corresponding dual space will be denoted by W −1,p (Ω), where + p10 = 1. N

Remark that for each u ∈ W01,p (Ω), ∇u ∈ (Lp (Ω)) and |∇u|p−2 ∇u ∈ 0 N Lp (Ω) . So, the operator p-Laplacian −∆p may be seen acting from W01,p (Ω) into its dual by Z | ∇u |p−2 ∇u∇w, for u, w ∈ W01,p (Ω). h−∆p u, wi = Ω

It’s well know (see e.g. El Khalil et al.9 ) that the inverse of the principal eigenvalue of p-Laplacian is the best constant in the Poincar´e’s inequality kukp ≤ ck∇ukp , ∀u ∈ W01,p (Ω). we can see also that ∆p is the unique duality mapping corresponding to the normalization numerical function f (t) = tp−1 with respect to the norm k∇.kp . Consequently the p-Laplacian is an homeomorphism between 0 W01,p (Ω) and W −1,p (Ω), with inverse is monotone; bounded and continuous. 2.2. Definitions In this article, all solutions are weak ones, i.e, (u, v) ∈ W01,p (Ω) × W01,q (Ω) is a solution of Sp,q (λ) if for all (φ, ψ) ∈ C0∞ (Ω) × C0∞ (Ω), R R R α−1 p−2 |v|β+1 uφ + µ Ω |u|α−1 |v|β+1 uφ  Ω |∇u| ∇u∇φ = λ Ω a(x)|u| R

Ω

|∇v|q−2 ∇v∇ψ = λ

R

Ω

a(x)|u|α+1 |v|β−1 vψ + µ

R

Ω

|u|α+1 |v|β−1 vψ.

In order to study the eigenvalue problem Sp,q (λ), we introduce some functional. For (u, v) ∈ W01,p (Ω) × W01,q (Ω), λ ∈ IR, we consider Z Z Z α+1 β+1 p q A(u, v) = |∇u| + |∇v| − λ a(x)|u|α+1 |v|β+1 ; p q Ω Ω Ω Z |u|α+1 |v|β+1 . B(u, v) = Ω

We set

and

n o M = (u, v) ∈ W01,p (Ω) × W01,q (Ω)/B(u, v) = 1 µ1 (λ) = inf {A(u, v)/(u, v) ∈ M} ,

(2.1)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

46

and for k ∈ IN ∗ Γk = {A ⊂ M : A is symmetric, compact, γ(A) = k} , where γ(A) is the genus of A, i.e, the smallest integer k such that there exists an odd continuous map from A to IRk \{0}, see Rabinowitz15 and Szulkin16 for details of genus. Definition 2.1. T is said to be in the (S+ ) class, if for any sequence un weakly convergent in X to u and lim sup < T un , un − u >≤ 0, it follows n→+∞

that un → u strongly inX. Definition 2.2. The principal eigencurve of Sp,q (λ) is the graph of the numerical function µ1 : λ −→ µ1 (λ) from IR into IR. 2.3. Auxiliary lemmas Now, we establish the following lemmas that we need in the proof of Theorem 3.1. Lemma 2.1. For all λ ∈ IR, we have (i) M is a closed C 1 -manifold. (ii) A and B are two even functionals and C 1 -differentiable in W01,p (Ω) × W01,q (Ω). (iii) A is bounded from below in M. Proof. Recall the following statements: (i) B is submersion C 1 -differentiable and B 0 (u, v) 6= 0 ∀(u, v) ∈ M (because (B 0 (u, v), (u, v)) = α + β + 2 6= 0). (ii) It is obviously. (iii) By H¨ older’s inequality we have for all (u, v) ∈ M Z Z Z α+1 β+1 A(u, v) = |∇u|p + |∇v|q − λ a(x)|u|α+1 |v|β+1 p q Ω Ω Ω Z Z β+1 α+1 p |∇u| + |∇v|q − |λ|kak∞ B(u, v) ≥ p q Ω Ω Z Z α+1 β+1 ≥ |∇u|p + |∇v|q − |λ|kak∞ p q Ω Ω ≥ λ1 − |λ|kak∞ , (2.2) where λ1 = µ1 (0). Thus A is bounded from below.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

47

Lemma 2.2. For all λ ∈ IR we have

i) B 0 is completely continuous in W01,p (Ω) × W01,q (Ω) and B is continuous for the weakly convergence. ii) A0 maps the bounded sets in the bounded sets. iii) If (un , vn ) * (u, v) weakly in W01,p (Ω) × W01,q (Ω) and A0 (un , vn ) converges strongly in (W01,p (Ω) × W01,q (Ω))0 , then (un , vn ) −→ (u, v) strongly in W01,p (Ω) × W01,q (Ω). iv) The restriction of A to M satisfies the Palais-Smale condition i.e, for (un , vn ) ∈ M if A(un , vn ) is bounded and (A|M )0 (un , vn ) −→ 0.

(2.3)

Then (un , vn ) has strong convergent subsequence in W01,p (Ω)×W01,q (Ω). Here (W01,p (Ω)×W01,q (Ω))0 is the dual space of W01,p (Ω)×W01,q (Ω) with respect to the standard product norm. Proof. i) and ii) are obviously. iii) Let (un , vn ) * (u, v) weakly in W01,p (Ω)×W01,q (Ω). and A0 (un , vn ) con∂A verges strongly in (W01,p (Ω)×W01,q (Ω))0 . Thus ∂A ∂u (un , vn ) (resp. ∂v (un , vn )) −1,p0 −1,q 0 converges strongly in W (Ω) (resp.W (Ω)). On the other hand we have 1 < ∂A < −∆p un , un − u >= α+1 ∂u (un , vn ), un − u > α−1 β+1 +λ < a|un | un |vn | , un − u >

and 1 < −∆q vn , vn − v >= β+1 < ∂A ∂v (un , vn ), vn − v > α+1 β−1 +λ < a|un | |vn | vn , vn − v > .

By passing to the limit we have lim sup < −∆p un , un − u > ≤ 0 and lim sup < −∆q vn , vn − v >≤ 0. n→+∞

n→+∞

Since p-Laplacian and q-Laplacian satisfy the condition (S + ) we conclude that W01,p (Ω)

strongly in × iv) From (2.3) we have A0 (un , vn ) −

(un , vn ) −→ (u, v)

W01,q (Ω)

as n → +∞.

< A0 (un , vn ), (un , vn ) > 0 B (un , vn ) −→ 0 as n → +∞, (2.4) α+β+2

and thanks to (2.2) we have the boundness of {(un , vn )}n in M. Therefore, there is a subsequence of (un , vn ) still denoted (un , vn ), weakly

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

48

convergent in W01,p (Ω) × W01,q (Ω). Moreover, by i) we have B 0 (un , vn ) → B 0 (u, v) strongly and B(u, v) = 1 (because B(un , vn ) = 1) , then (u, v) ∈ M; and by ii) we have (A0 (un , vn ))n is bounded. This and (2.4) implies that A0 (un , vn ) converges strongly in (W01,p (Ω) × W01,q (Ω))0 . Finally, by iii), we conclude that (un , vn ) → (u, v) strongly in W01,p (Ω) × W01,q (Ω). 3. Main results Theorem 3.1. For any λ ∈ IR and any integer k ∈ IN ∗ , µk (λ) = inf

max A(u, v)

A∈Γk (u,v)∈A

(3.1)

is critical value of A(u, v) in M. More precisely, there exist (uk (λ), vk (λ)) ∈ M, µk (λ) ∈ IR such that, A(uk (λ), vk (λ)) = µk (λ). With (uk (λ), vk (λ)) is the eigenfunction of Sp,q (λ) for the eigenpair eigenvalue (λ, µk (λ)). Proof. From Szulkin16 and using Lemma 2.1 and Lemma 2.2 we need only to prove that for all k in IN ∗ Γk 6= ∅. Indeed, W01,p (Ω) × W01,q (Ω) is separable, thus for all k ∈ IN ∗ there exist e1 , e2 , ...., ek k functions of W01,p (Ω) × W01,q (Ω) linearly dense in W01,p (Ω) × W01,q (Ω), where ei = (eip , eiq ) with suppei ∩ suppej = ∅ for all i 6= j and B(ei ) = 1. Denote Fk = span(e1 , e2 , ...., ek ) is subspace of W01,p (Ω) × W01,q (Ω) and dimFk = k; then if v ∈ Fk there exist t1 , t2 , ...., tk real numbers such that i=k α+β+2 v = Σi=k . i=1 ti ei thus B(v) = Σi=1 |ti | 1 Therefore the map v ∈ Fk −→ B(v) α+β+2 is a norm in Fk . Let k(., .)kp,q the norm in W01,p (Ω) × W01,q (Ω); hence, there is c > 0 such that for all v ∈ Fk , we have 1 1 ckvkp,q ≤ B(v) α+β+2 ≤ kvkp,q . c This implies that the set n o V = Fk ∩ (u, v) ∈ W01,p (Ω) × W01,q (Ω)/B(u, v) ≤ 1 6= ∅

is a bounded neighborhood symmetric of (0, 0) ∈ Fk . Thus by (f ) of Prop. 2.3 of Szulkin,16 γ(Fk ∩ M) = k then Γk 6= ∅.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

49

Corollary 3.1. For all λ ∈ IR we have lim µk (λ) = +∞

k→+∞

Proof. W01,p (Ω) × W01,q (Ω) is separable, hence we can have bi-orthogonal system (ek , e∗n )n such that ek ∈ W01,p (Ω) × W01,q (Ω), e∗n ∈ (W01,p (Ω) × W01,q (Ω))0 with (ek )k is linearly dense in W01,p (Ω) × W01,q (Ω); and e∗j are total in (W01,p (Ω) × W01,q (Ω))0 , see e.g Szulkin.16 Set for k ∈ IN ∗ , Fk = span(e1 , e2 , . . . , ek ), Fk⊥ = span(ek+1 , ek+2 , ek+3 . . .);

Therefore, by (g) of proposition 2.3 in Szulkin,16 for all A ∈ Γk we have ⊥ A ∩ Fk−1 6= ∅ ∀k ≥ n hence mk = inf

sup A(u, v) −→ +∞ as n → +∞

A∈Γk A∩F ⊥ k−1

(3.2)

⊥ indeed, if not, for k large enough, there is (uk , vk ) ∈ Fk−1 such that B(uk , vk ) = 1 and

mk ≤ A(uk , vk ) ≤ m, for some constant m independent of k. Thus Z Z Z α+1 p β +1 q |∇uk | + |∇vk | −λ a(x)|uk |α+1 |vk |β+1 ≤ m. A(uk , vk ) = p Ω q Ω Ω

This implies

α+1 p

Z

Ω

|∇uk |p +

β+1 q

Z

Ω

|∇vk |q ≤ m + |λ|kak∞ .

Hence (uk , vk ) is bounded in W01,p (Ω) × W01,q (Ω). Since (ek , e∗n ) = 0 ∀k ≥ n and by the Sobolev’s imbedding we have (uk , vk ) −→ (0, 0) as n → +∞, strongly in Lp (Ω) × Lq (Ω). This is a contradiction, because B(uk , vk ) = 1. Since µk (λ) ≥ mk so we conclude by (3.2) that lim µk (λ) = +∞.

k→+∞

Corollary 3.2. For all λ fixed in IR µ1 (λ) given by (2.1) is the smallest eigenvalue of Sp,q (λ) and there exists (u, v) ∈ M a solution of Sp,q (λ) associated for (λ, µ1 (λ)). Proof. Let (u, v) ∈ M and B = {(u, v), (−u, −v)} therefore γ(B) = 1 and B ∈ Γ1 . Since A is even then µ1 (λ) ≤ inf (u,v)∈M A(u, v), the reverse inequality is obvious.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

50

Lemma 3.1. i) If (u, v) is a solution of (Sp,q (λ)) then (u, v) ∈ L∞ (Ω) × L∞ (Ω). ii) For all λ ∈ IR let (u, v) be an eigenvector associated of µ1 (λ) such that 1,η u ≥ 0, v ≥ 0 and B(u, v) = 1 then: u > 0 , v > 0 and (u, v) ∈ Cloc (Ω)× 1,η Cloc (Ω). ´lin.6 Proof. i) deduced from the results of The ii) Let (u, v) be a solution of (Sp,q (λ)) associated of (λ, µ(λ)). Thus (|u|, |v|) is also a solution of (2.1). Indeed, A(|u|, |v|) ≤ A(u, v) and B(|u|, |v|) = 1. Hence, we can suppose that u ≥ 0 and v ≥ 0, on the other hand by (i) and Strong Maximum Principle of Vazquez17 we confirm that u > 0 , v > 0. The results of regularity follow from (i) and local regularity of Dibinedetto.5 Now, let p−2 p Z Z Z |u| u |u| −p |∇φ|p−2 ∇φ∇u Γp (u, φ) = |∇u|p +(p−1) |∇φ|p φ φp−1 Ω Ω Ω 2 for all (u, φ) ∈ W01,p (Ω) with φ > 0 in Ω.

Lemma 3.2. For all (u, φ) ∈ (W01,p (Ω) ∩ C(Ω))2 with φ > 0 in Ω, we have Γp (u, φ) ≥ 0 and if Γp (u, φ) = 0 there is c ∈ IR such that u ≡ cφ. Proof. By Young’s inequality, we have for > 0, p−2

p−1 u p−1 (φ) ∇u|∇φ|p−2 ∇φ u|u| φp−1 ≤ |∇u||∇φ| u p p p ≤ p |∇u| + p−1 pp | φ | |∇φ|.

(3.3)

For = 1 we have by integration over Ω, Z Z Z u u p |∇φ|p−2 ∇φ∇u( )p−1 ≤ |∇u|p + (p − 1) | |p |∇φ|p , φ φ Ω Ω Ω

thus

Γp (u, φ) ≥ 0. On the other hand, if Γp (u, φ) = 0 by (3.3), we obtain Z Z Z |u|p−2 u u p−2 p p |∇φ| ∇φ∇u( p−1 ) − |∇u| − (p − 1) | |p |∇φ|p = 0 (3.4) φ φ Ω Ω Ω and

Z Ω

u|u|p−2 ∇u∇φ|∇φ|p−2 p−1 φ

− |∇u||∇φ|

p−1

u p−1 dx = 0. ( ) φ

(3.5)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

51

By (3.4) we find |∇u| = | φu ∇φ| thus from (3.5), it follows that ∇u = φu ∇φ, where || = 1. Hence Γp (u, φ) = 0 implies = 1 and ∇( uφ ) = 0. Therefore, there is c ∈ IR such that u = cφ. λ1 −λ1 , 2kak ] , if (u(λ), v(λ)) is the eigenvecTheorem 3.2. For all λ ∈ [ 2kak ∞ ∞ tor corresponding to µ1 (λ)satisfying B(u, v) = 1, u > 0, v > 0 then (u, v) is unique.

Proof. Let (u, v) , (φ, ψ) be two positive eigenvectors associated with µ1 (λ) therefore, −∆p u = λa(x)|u|α−1 |v|β+1 u + µ|u|α−1 |v|β+1 u (a) α+1 β−1 α+1 β−1 −∆q v = λa(x)|u| |v| v + µ|u| |v| v (b) and −∆p φ = λa(x)|φ|α−1 |ψ|β+1 φ + µ|φ|α−1 |ψ|β+1 φ −∆q ψ = λa(x)|φ|α+1 |ψ|β−1 ψ + µ|φ|α+1 |ψ|β−1 ψ.

(c) (d)

For any > 0, let φ (u, φ) =

up vq ψ (v, ψ) = . (φ + )p−1 (ψ + )q−1

Hence, multiplying (a) by u and (c) by φ (u, φ), integrating by parts over Ω, and taking the difference, we obtain Z u v u Γp (u, φ) = (λa(x) + µ1 )|φ|α+1 |ψ|β+1 [| |α+1 | |β+1 − | |p ], (3.6) φ ψ φ Ω multiplying (b) by v and (d) by ψ (v, ψ), integrating by parts over Ω and taking the difference, we obtain Z v v u Γq (v, ψ) = (λa(x) + µ1 )|φ|α+1 |ψ|β+1 [| |α+1 | |β+1 − | |p ]. (3.7) φ ψ ψ Ω So, multiplying (3.6) by

α+1 p

and (3.7) by

β+1 α+1 Γp (u, φ)+ Γq (v, ψ) = p q −

Z

β+1 q ,

we have

u v (λa(x)+µ1 )|φ|α+1 |ψ|β+1 [| |α+1 | |β+1 φ ψ Ω

α+1 u p β+1 v q | | − | | ]. p φ q ψ

By Young’s inequality we have u v α+1 u p β+1 v q | |α+1 | |β+1 − | | − | | ≤ 0. φ ψ p φ q ψ

(3.8)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

52

According to Lemma 3.2 and (3.8), we have by using |λ| ≤ 0≤

λ1 2kak∞ ,

β+1 α+1 Γp (u, φ) + Γq (v, ψ) ≤ 0. p q

Hence Γp (u, φ) = Γq (v, ψ) = 0. This and Lemma 3.2 imply that φ ≡ tu and ψ ≡ t0 v. Then by using the normalization B(u, v) = 1 we conclude that u ≡ φ, v ≡ ψ. Acknowledgements. The author gratefully acknowledges the financial support provided by Al-Imam Muhammed Ibn Saud Islamic University during this research. References 1. A. Anane: Etude des valeurs propres et de la r´esonance pour l’op´erateur p-Laplacien, th`ese de Doctorat, U.L.B (1987-88). 2. P.A. Binding and Y.X. Huang: The principal eigencurve for p-Laplacian, Diff. int. Equations, 8, n.2 (1995), 405-415. 3. J. Chabrowski: On multiple solutions for nonhomogeneous system of elliptic equations, Revista Matimatica de la Universidad Computense de Madrid volume 9, numero 1, 1996. 4. J.I. Diaz and J.E. Saa: Existence et unicit´e de solutions positives pour certaines ´equations elliptiques quasilin´eaires. C.R. Acad. Sci. Paris, serie I, Math. 305,(1987), 521-524. 5. E. Dibenedetto: C 1+α -local regularity of weak solutions of degenerate elliptic equations, Nonlinear Analysis T.M.A. 7, (1983), 827-859. 6. F. De Th´ elin: Premi`ere valeur propre d’un syst`eme elliptique non lin´eaire, C.R. Acad. Sci. Paris,t. 311, Serie I, (1990), 603-606. 7. F. De Th´ elin and J. V´ elin: Existence et non existence de solutions non triviales pour des syst´emes elliptiques non-lin´eaires, C.R. Acad. Sci. Paris, 313, serie I, (1991), 589-592. 8. A. El Khalil, A. El Manouni and M. Ouanan: Simplicity and stability of the first eigenvalue of a nonlinear elliptic system, International Journal of Mathematics and Mathematical Sciences, Vol. (2005)10, (2005), 1555-1563. 9. A. El Khalil and A. Touzani: On the first eigencurve of the p-Laplacian, Lecture Note in Pure and Applied Mathematics, Vol. 229 (2002), 195-205. 10. J. Fleckinger, R.F. Manasevich, N.M. Stavrakakis and F. De Th´ elin: Principal eigenvalue for some quasilinear elliptic equations on IR N , Advances in Diffrential Equations, Vol. 2, No.6 (1997), 981-1003. 11. P. Hess and T. Kato: On some liner and nonliner eigenvalue problems with an indefinite weight function, Comm.P.D.E. 5 (1980), 999-1030. 12. T. Kato: Superconvexity of the spectral radius and convexity of the spectral bond and the type, Math.Z., 180 (1982), 265-273.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

53

13. P. Lindqvist: On the equation div(|∇u|p−2 ∇u)+λ|u|p−2 u = 0, Proc.Amer. Math. Soc., 109 (1990), 157-164. 14. R.G.D. Richardson: Theorems of oscillation for tow linear differential equations of the second order with two parameters, Tran. Amer. Math. SOC., 13(1992), 22-34. 15. P.H. Rabinowitz: Minimax methods in critical point theory with applications to differential equations, AMS Rigional Conference Series in Math. vol. 65 (1986). 16. A. Szulkin: Ljusternik-Schnirelmann theory on C 1 -manifolds, Ann. Inst. Henri Poincar´e, Anal. Non. 5 (1988), 119-139. 17. J.L. Vazquez: A strong maximum principle for some quasilinear elliptic equations, Appl. Math. Optim. 12 (1984), 191-202.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

54

Parabolic inclusions with nonlocal conditions Abdelkader Boucherif King Fahd University of Petroleum and Minerals Department of Mathematics and Statistics Box 5046 Dhahran, 31261, Saudi Arabia E-mail: [email protected] In this paper we consider the solvability of nonlinear parabolic differential equations with discontinuous nonlinearities, subjected to nonlocal conditions. We are concerned with the existence of solutions. Our technique is based on the Green’s function for linear parabolic partial differential equations, the maximum principle, fixed point theorems for multivalued maps and the method of lower and upper solutions. Keywords: Parabolic problems; Integral representation of solutions; Maximum principles; Multivalued maps; Nonlocal conditions; Fixed point theorems; Lower and upper solutions.

1. Introduction Let Ω be an open bounded domain in RN , N ≥ 2, with a smooth boundary ∂Ω. Let T be a positive real number, D = Ω × (0, T ] and Γ = ∂Ω × [0, T ]. Given continuous functions φ, βi : Ω → R, i = 1, 2, ..., m and numbers ti , i = 1, 2, ..., m in (0, T ) with 0 < t1 < ... < tm < T, we are concerned with the existence of solutions of the following parabolic problem with a multivalued right-hand side and nonlocal conditions Dt u + Lu ∈ F (x, t, u) u(x, t) = 0

u(x, 0) +

m X i=1

(x, t) ∈ D, (x, t) ∈ Γ,

βi (x)u(x, ti ) = φ (x)

(1)

(2)

x ∈ Ω,

(3)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

55

where L is a strongly elliptic operator given by Lu = −

N X

aij (x, t)Di Dj u + c(x, t)u.

i,j=1

Parabolic problems with discontinuous nonlinearities arise in the description of many phenomena in the applied sciences. We can mention, for instance, chemical reactor theory,15 porous medium combustion,13 ,14 best response dynamics arising in game theory,10 ,18 . Parabolic problems with discontinuous nonlinearities have been also investigated in the papers,6 ,8 ,7 ,30 ,31 ,33 . The importance of nonlocal conditions and their applications in different field has been discussed in,3 ,5 ,12 . Several papers have dealt with parabolic problems with continuous nonlinearities and nonlocal conditions. See for instance,20 ,21 ,25 ,26 ,34 ,35 . In this paper we consider a nonlocal problem for a class of nonlinear parabolic equations with a multivalued right hand side. We shall convert problem (1), (2), (3), to an integral inclusion using the properties of the Green’s function corresponding to the linear problem. We, then, provide sufficient conditions on the data that will enable us to obtain at least one solution. Our approach is based on fixed point theorems for suitable multivalued operators and the method of lower and upper solutions. The outline of the paper is as follows. In section 2 we introduce notations and function spaces which will be used in the paper. In section 3, we shall study the linear nonhomogeneous problem and the properties of the Green’s function. In section 4 we recall the main properties of multivalued maps. We state and prove our main results in section 5.

2. Preliminaries In this section we introduce some notations and function spaces. Let Ω be an open bounded domain in RN , N ≥ 2, with a smooth boundary ∂Ω. Let T be a positive real number, D = Ω × (0, T ) and Γ = ∂Ω × [0, T ]. Then Γ is smooth and any point on Γ satisfies the inside (and outside) strong sphere property, see;16 i.e. for any (x0 , t0 ) ∈ Γ there is a closed ball B ⊂ Ω e outside Ω) such that Γ ∩ (B × [0, T ]) = {(x0 , t0 )}, (and a closed ball B e (and Γ ∩ B × [0, T ] = {(x0 , t0 )}). For u : D → R we denote its partial

derivatives (when they exists) by Dt u = ∂u/∂t, Di u = ∂u/∂xi , Di Dj u = ∂ 2 u/∂xi ∂xj , i, j = 1, ..., N. C(D) denotes the Banach space of continuous functions u : D → R,

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

56

endowed with the norm |u|0 = sup{|u(x, t)| ; (x, t) ∈ D}. We say that u ∈ C 2,1 (D) if u, Di u, Di Dj u and Dt u exist and are continuous on D. In fact, we can write C 2,1 (D) = {u ∈ C(D); u(., t) ∈ C 2 (Ω), t ∈ (0, T ) , u(x, .) ∈ C 1 (0, T ) , x ∈ Ω}. u ∈ C(D) is called H¨ older continuous of order α ∈ (0, 1] if

|u(x, t) − u(ξ, τ )| Hα (u) = sup{ α/2 ; (x, t) , (ξ, τ ) ∈ D} < +∞. 2 2 kx − ξk + |t − τ |

In this case we write u ∈ C α (D) and we define its norm by |u|α = |u|0 + Hα (u) .

If α = 1, u is called Lipschitz continuous and we write u ∈ Lip(D). Note that the natural injection i : C α (D) → C(D) is continuous. We say that that C α (D) is continuously embedded in C(D), and we write C α (D) ,→ C(D). Also, u ∈ C 2+α,1+α (D) if u(., t) ∈ C 2+α (Ω) for all t ∈ (0, T ) and u(x, .) ∈ C 1+α (0, T ) for all x ∈ Ω. For u ∈ C 2+α,1+α (D) we define its norm by |u|2+α,1+α = |u|α +

N X i=1

|Di u|α +

N X

i,j=1

|Di Dj u|α + |Dt u|α .

We say that ∂Ω is in the class C `+α , ` ∈ N, α ∈ [0, 1) if in a neighborhood of each point of ∂Ω there is a local reprentation of ∂Ω having the form xi = ϑi (x1 , . . . , xi−1 , xi+1 , . . . , xN ) with ϑi ∈ C `+α . Also, for 1 ≤R p < +∞, we say that u : D → R is in Lp (D) if u is p measurable and D |u(x, t)| dxdt < +∞, in which case we define its norm by 1/p Z p |u(x, t)| dxdt . |u|Lp = D

3. Linear nonhomogeneous problem In this section we consider the linear nonhomogeneous problem Dt u + Lu = f (x, t), u(x, t) = 0,

(x, t) ∈ D, (x, t) ∈ Γ,

(4) (5)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

57

with the following nonlocal initial condition u(x, 0) +

m X

βi (x)u(x, ti ) = φ (x) ,

i=1

x ∈ Ω.

(6)

We shall assume throughout this paper that the function φ : Ω → R is continuous and the functions aij , c : D → R are H¨ older continuous, aij = aji and moreover there exist positive numbers λ0 , λ1 such that 2

λ0 kξk ≤

N X

i,j=1

2

aij (x, t)ξi ξj ≤ λ1 kξk , ∀ξ ∈ RN and ∀ (x, t) ∈ D.

The classical problem, i.e.when βi = 0 for all i = 1, 2, ..., m , is well known and completely solved (see the books,16 ,22 ,24 ,28 ). Problem (4), (5), (6) has been investigated by several authors (see for instance,9 ,21 ,35 and the references therein). The following version of the maximum principle can be found in,9 ,28 . Lemma 3.1. Let u ∈ C 2,1 (D) ∩ C(D). Assume that c(x, t) ≥ c0 > 0 on D Pm and βi (x) ≤ 0 on Ω with −1 ≤ i=1 βi (x) ≤ 0 on Ω. If Dt u + Lu ≥ 0 in Pm D, u(x, t) ≥ 0 on Γ, u(x, 0)+ i=1 βi (x)u(x, ti ) ≥ 0 on Ω. Then u(x, t) ≥ 0 on D. Moreover, either u(x, t) = 0 ∀ (x, t) ∈ D, or u(x, t) > 0 on D. Proof. Suppose there exists (η, τ ) ∈ D such that u (η, τ ) < 0. It follows from the continuity of u that u achieves a negative minimum at some point (x0 , t0 ) ∈ D. By the strong maximum principle (see16 ) we have either (x0 , t0 ) ∈ Γ or (x0 , t0 ) = (x0 , 0) with x0 ∈ Ω. From the above assumptions P we see that minD u (x, t) = u (x0 , 0) < 0. Hence, if 1 + m i=1 βi (x0 ) > 0, ! m m X X 0 ≤ u(x0 , 0) + βi (x0 )u(x0 , ti ) ≤ u(x0 , 0) 1 + βi (x0 ) < 0, i=1

i=1

which is a contradiction. Pm In case 1 + i=1 βi (x0 ) = 0 we let u(x0 , tk ) = min{u (x0 , ti ) ; i = 1, 2, ..., m}. Then u(x0 , 0) − u(x0 , tk ) = u(x0 , 0) + u(x0 , tk ) ≥ u(x0 , 0) +

m X i=1

m X

βi (x0 )

i=1

βi (x0 )u (x0 , ti ) ≥ 0.

This last inequality shows that u takes on a negative minimum at (x0 , tk ) ∈ D, which is not possible. This completes the proof of the Lemma.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

58

One of the important features of the linear problem is the integral reprentation of solutions. The homogeneous problem (x, t) ∈ D

Dt u + Lu = 0,

(x, t) ∈ Γ

u(x, t) = 0, u(x, 0) +

m X

βi (x)u(x, ti ) = 0,

i=1

x∈Ω

P has a only the trivial solution, provided that −1 ≤ m i=1 βi (x) ≤ 0. There exists a unique function, G(x, t; y, s), called the Green’s function corresponding to the linear homogeneous problem. This function satisfies the following (,16 ,2228 ), (i) Dt G + LG = δ (t − s) δ (x − y), s < t, x, y ∈ Ω (ii) G(x, t; y, s) = 0, s > t, x, y ∈ Ω (iii) G(x, t; y, s) = G(x, t; y, s) = 0, s < t, x, y ∈ ∂Ω, (iv) G(x, t; y, s) > 0 for (x, t) ∈ D (v) G, Dt G, Di G, Di Dj G are continuous functions of (x, t), (y, s) ∈ D, t − s > 0, ! 2 −a kx − yk −N/2 , for some posi(vi) |G(x, t; y, s)| ≤ C (t − s) exp t−s tive constants C, a. Lemma 3.2. Assume that the function f is H¨ older continuous and Pm |β (x)| < 1. Then, Problems (4), (5), (6) has a unique strong soi i=1 2,1 lution, i.e. a solution u ∈ C (D) ∩ C(D). Proof. Consider the following representation (,9 ,1628 ),

u(x, t) =

Z

G(x, t; y, 0) u (y, 0) dy + Ω

Z tZ 0

G(x, t; y, s) f (y, s) dyds

(7)

Ω

for (x, t) ∈ D, where u (y, 0) has to be determined. Using condition (6) we see that u(x, 0) is a solution of the following Fredholm integral equation of the second kind R Pm   u(x, 0) + i=1 βi (x) Ω G(x, ti ; y, 0) u (y, 0) dy = 

−

Pm

i=1 βi (x)

R ti R 0

G(x, ti ; y, s) f (y, s) dyds + φ (x) . Ω

(8)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

59

Pm The condition i=1 |βi (x)| < 1 implies that Eq. (8) with f = 0 and φ = 0 has only the trivial solution. Hence there exists a unique solution u(., 0) of (8). Consequently (7) gives the unique solution of (4), (5), (6). From the above discussion we see that for each v ∈ C (Ω) the problem (4), (5) and u(x, 0) = v(x)

(9)

has a unique solution u ∈ C 2,1 (D) ∩ C(D), which we denote by u(x, t; v) and has the following representation Z Z tZ G(x, t; y, 0) v (y) dy G(x, t; y, s) f (y, s) dyds + u(x, t; v) = 0

Ω

Ω

for (x, t) ∈ D. It is clear from the above representation Rthat the functions ϕ tR and ψ defined Rrespectively by ϕ (x, t) = G(x, t; y, s) dyds 0 Ω and ψ (x, t) = Ω G(x, t; y, 0)dy are continuous on D. Let K0 := RtR sup(x,t) 0 Ω G(x, t; y, s) dyds. Following26 we define an operator Υ : C (Ω) → C (Ω) , by Υv (x) = −

m X

βi (x)u(x, ti ; v) + φ (x) .

(10)

i=1

Lemma 3.3. Assume maxx∈Ω { Υ is a Lipschitz operator.

Pm

i=1

|βi (x)|

R

Ω

G(x, ti ; y, 0)dy} < 1. Then

Proof. Let v1 , v2 ∈ C (Ω) and let u1 (x, t; v1 ) and u2 (x, t; v2 ) be the corresponding solutions of (4), (5), (9) respectively.

(Υv1 )(x, t) − (Υv2 )(x, t) = −[

m X i=1

βi (x)u(x, ti ; v1 ) −

m X

βi (x)u(x, ti ; v2 )].

i=1

It follows from (7) that (Υv1 )(x, t) − (Υv2 )(x, t) = −

m X i=1

βi (x)

Z

Ω

G(x, ti ; y, 0) [v1 (y) − v2 (y)]dy.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

60

Thus |(Υv1 )(x, t) − (Υv2 )(x, t)| Z m X ≤{ |G(x, ti ; y, 0)| dy} kv1 − v2 k0 . |βi (x)| i=1

Ω

Therefore k(Υv1 ) − (Υv2 )k0 ≤ maxx∈Ω {

m X i=1

|βi (x)|

Z

Ω

|G(x, ti ; y, 0)| dy} kv1 − v2 k0 .

Consequently, Υ has a unique fixed point v0 ∈ C (Ω) , and u = u(x, t; v0 ) is the unique solution of (4), (5), (6). 4. Multivalued functions In this section we introduce some useful definitions and properties from set-valued analysis. For complete details on multivalued maps we refer the interested reader to the books,1 ,2 ,11 and.19 Let (X, |.|X ) and (Y, |·|Y ) be Banach spaces. The domain of a multivalued map < : X → 2Y is the set dom< = {z ∈ X; <(z) 6= ∅}. < is convex (closed) valued if <(z) is convex (closed) for each z ∈ X. < is bounded on bounded sets if < (A) = ∪z∈A < (z) is bounded in Y for all bounded subsets A ⊂ X (i.e. supz∈A {sup{|y| ; y ∈ <(z)}} < ∞). < is called upper semicontinuous (usc) on X if for each z ∈ X the set <(z) is nonempty, closed in Y , and for each open subset B of Y containing <(z), there exists an open neighborhood A of z in X such that <(A) ⊂ B. In terms of sequences, < is usc if for each sequence (zn ) ⊂ X, zn → z0 , and B a closed subset of Y such that <(zn ) ∩ B 6= ∅ then <(z0 ) ∩ B 6= ∅. The set-valued map < is called completely continuous if <(A) is relatively compact in Y for every bounded subset A of X. If < is completely continuous with nonempty compact values, then < is usc if and only if < has a closed graph (i.e. zn → z, wn → w, wn ∈ <(zn ) ⇒ w ∈ <(z)). < is called lower semicontinuous (lsc) on X if <−1 (B) is open in X whenever B is open in Y , or {z ∈ X; < (z) ⊂ B} is closed in X whenever B is closed in Y . It can be shown that < is lsc if and only if dY (y, <(.)) is usc for every y ∈ Y. When X ⊂ Y then < has a fixed point if there exists z ∈ X such z ∈ <(z). When D ⊂ X = RN +1 and Y = R the multivalued map < : D → 2R with closed values is called measurable if for every κ ∈ R, the function v 7−→ dist(κ, <(v)) = inf{|κ − z| ; z ∈ <(v)} is measurable.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

61

Definition 4.1. A multivalued map F : D × R → 2R \∅ is called an L2 − Carath´eodory multifunction if (i) (x, t) 7→ F (x, t, u) is measurable for each u ∈ R, (ii) u 7→ F (x, t, u) is upper semicontinuous for almost all (x, t) ∈ D, (iii) for each r > 0 there exists ωr ∈ L2 (D) such that |F (x, t, u)| ≤ ωr (x, t) a.e. on D whenever |u| ≤ r. Definition 4.2. For u ∈ C(D), the set of L2 −selections of the multivalued map F is defined by 2 SF,u = {v ∈ L2 (D); v(x, t) ∈ F (x, t, u(x, t)), a.e. (x, t) ∈ D}.

This set is not empty ( [23, Lemma 3]). Definition 4.3. Let F : D × R → 2R have nonempty compact values. The 2 Nemitsky operator of F is the set-valued operator F : C(D) → 2L (D) , defined by F(u) := {w : D → R measurable; w(x, t) ∈ F (x, t, u(x, t)), a.e. (x, t) ∈ D}. Theorem 4.1. (,2732 ). Let F be an L2 − Carath´eodory multifunction. Then the Nemitsky operator F is weakly completely continuous and integrable on bounded sets. Using the properties of the Green’s function we get the following results (27 ). Lemma 4.1. Assume the single valued map g ∈ C(D). Then the operator γ : Lip(D) → C(D), defined by Z tZ G(x, t; y, s)f (y, s)dyds γf (x, t) = g(x, t) + 0

Ω

is continuous. Lemma 4.2. Assume F : D × R → R is an L2 − Carath´eodory multifunction with nonempty, compact, convex values, with a Lipschitz continuous selection, and g ∈ C(D). Then the operator γ ◦ F is of usc type; i.e. is usc, completely continuous and has nonempty, compact, convex values. Theorem 4.2. (Nonlinear alternative for multivalued maps) Let K be a convex subset of a Banach space E, U ⊆ K be relatively open, and 0 ∈ U. Suppose Λ : U → K is an usc compact multivalued map with nonempty, compact, convex values. Then either

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

62

(i) there is u ∈ U such that u ∈ Λu; or (ii) there is u ∈ ∂U and a λ ∈ (0, 1) such that u ∈ λΛu. The above theorem can also be found in.17 Definition 4.4. Let (Z, d) be a metric space and let A, B be two nonempty subsets of Z. The Hausdorff distance between A and B is defined by dH (A, B) := max{sup d(a, B), supd(A, b)}, a∈A

b∈B

where d(a, B) = inf{d(a, b); b ∈ B} and d(A, b) = inf{d(a, b); a ∈ A}. Then one can show that (Pcl,b (Z), dH ) is a metric space. Here Pcl,b (Z) denotes the collection of all closed bounded subsets of Z. Definition 4.5. A multivalued operator Λ : Z → 2Z , with closed values is called (i) δ−Lipschitz if and only if there exists δ > 0 such that dH (Λ(u), Λ(v)) ≤ δ d (u, v) for all u, v ∈ Z (ii) a contraction if and only if it is δ−Lipschitz with δ < 1. N A subset Σ ⊂ D ×R is L B measurable if Σ belongs to the σ−algebra generated by all sets of the form D × J where D is Lebesgue measurable in D and J is Borel measurable in R. Definition 4.6. Let (Z, d) be a separable metric space. We say that the multivalued operator < : Z → P L1 (D; R) has property (BC) if < is lsc with nonempty, closed and decomposable values. Remark 4.1. < (z) is decomposable if for all u, v ∈ < (z) and measurable ∆ ⊂ D, we have uχ∆ + vχD−∆ ∈ < (z) , where χ∆ is the characteristic function of the set ∆. We state a selection theorem due to Bressan and Colombo (4 ) Theorem 4.3. Let (Z, d) be a separable metric space and let < : Z → P L1 (D; R) has property (BC). Then < has a continuous selection, i.e. there exists a single valued function ρ : Z → L1 (D; R) such that ρ (z) ∈ < (z) for every z ∈ Z. 5. Main results We shall be interested in strong solutions of problem (1), (2), (3). Definition 5.1. u ∈ C (D) is called a strong solution of (1), (2), (3) if there exists a single-valued function f ∈ Lip(D) such that f (x, t) ∈ F (x, t, u(x, t)) and (4), (5), (6) hold.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

63

5.1. Lipschitz multivalued right-hand sides Theorem 5.1. Suppose that F : D × R → 2R is a multifunction with compact values and the following conditions are satisfied. (H0) There exists f ∈ Lip(D) such that f (x, t) ∈ F (x, t, u(x, t)) Theorem 5.2. (H1) There exists ` ∈ Lip(D) such that dH (F (x, t, u) , F (x, t, z)) ≤ ` (x, t) |u − z| , a.e. (x, t) ∈ D, u, z ∈ R,

dH (0, F (x, t, 0) ≤ ` (x, t) a.e. (x, t) ∈ D, RtR (H2) max(x,t)∈D { 0 Ω G(x,Rt; ξ, τ )` (ξ, τ ) dξdτ < 1. Pm (H3) maxx∈Ω { i=1 |βi (x)| Ω G(x, ti ; y, 0)dy} < 1. Then Problem (1), (2), (3) has at least one solution. Proof. For v ∈ C(Ω) consider the problem (1), (2), (9) i.e. Dt u + Lu ∈ F (x, t, u), u(x, t) = 0,

(x, t) ∈ D,

u(x, 0) = v(x),

(x, t) ∈ Γ, x ∈ Ω.

u is a solution of this problem if and only if u ∈ X = C(D) is a fixed point of the multivalued operator < : X → 2X , defined by
Z tZ 0

Z

Ω

G(x, t; y, 0) v(y)dy, Ω

(11)

(x, t) ∈ D,

G(x, t; y, s) F(u) (y, s) dyds,

(x, t) ∈ D.

(12)

(13)

Notice that < is the sum of a single-valued operator gv and a multivalued operator GF, where F is the Nemitsky operator associated with the multifunction F (x, t, u).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

64

We show that
Ω

Since F has compact values, it follows that ( fn )n∈N , passing to subsequences if necessary, converges to some h ∈ Lip(D). Hence Z tZ G(x, t; y, s) h(y, s)dyds, (x, t) ∈ D, z(x, t) = gv (x, t) + 0

Ω

which shows that z ∈
Then z1 (x, t)−z2 (x, t) =

Ω

Z tZ 0

G(x, t; y, s) [h1 (y, s)−h2 (y, s)]dyds, Ω

(x, t) ∈ D.

(H1) yields |z1 (x, t) − z2 (x, t)| ≤

Z tZ 0

≤{

G(x, t; y, s)`(y, s) |u1 (y, s) − u2 (x, t)| dyds

Ω

Z tZ

G(x, t; y, s) `(y, s)dyds} ku1 − u2 k0 Z tZ ≤ max { G(x, t; y, s)` (y, s) dyds} ku1 − u2 k0 . 0

Ω

(x,t)∈D

0

Ω

This shows that dH (
Z tZ 0

G(x, t; y, s)` (y, s) dyds}. Ω

It follows from (H2) that < is a contraction. Theorem 11.1 in11 implies that < has at least one fixed point u0 . Then u0 satisfies Z Z tZ u0 (x, t) ∈ G(x, t; y, 0) v(y)dy + G(x, t; y, s) F (y, s, u0 (y, s))dyds Ω

0

Ω

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

65

for (x, t) ∈ D. Lemma 3.3 shows that u0 (x, t; v0 ) , where v0 is the unique fixed point of the operator Υ, which exists by (H3), is a solution of problem (1), (2), (3). 5.2. Upper semicontinuous right-hand sides Theorem 5.3. Let F : D × R → R be an L2 − Carath´eodory multifunction with nonempty, compact, convex values. Assume that, in addition to (H0) and (H3), the following conditions are satisfied. (H4) There exist q ∈ C(D) and Ψ : [0, ∞) → (0, ∞) continuous and nondecreasing such that |F (x, t, u| ≤ q(x, t)Ψ (|u|) for almost all (x, t) ∈ D and u ∈ R. ρ > 1. (H5) supρ∈[0,∞) |gv0 |0 + K0 |q|0 Ψ (ρ) Then problem (1), (2), (3) has at least one solution. Proof. Recall that u is a solution of (1), (2), (3) if and only if u is a fixed point of the multivalued operator < given by (11). Lemma 4.2 implies that < is of upper semi-continuous type. It follows from condition (H3) and the properties of the Green’s function that for v0 ∈ C(D), the unique fixed point of the operator Υ, the function gv0 is well defined and continuous. Let M0 > 0 satisfy M0 > 1. |gv0 |0 + K0 |q|0 Ψ (M0 ) This is possible because of (H5). Consider U := {u ∈ X; |u|0 < M0 }. Then U is relatively open in K = X = C(D). We shall apply Theorem 4.2 to the operator <, and show that the second alternative does not hold. Let u ∈ X be a solution of Z tZ u(x, t) ∈ λ(gv0 (x, t) + G(x, t; y, s) F (y, s, u (y, s))dyds), (x, t) ∈ D, 0

Ω

(14) with λ ∈ (0, 1) . There exists a function f ∈ Lip (D) such that f (x, t) ∈ F (x, t, u(x, t)). From (12) we obtain Z tZ G(x, t; y, s) f (y, s)dyds), (x, t) ∈ D. (15) u(x, t) = λ(gv0 (x, t) + 0

Ω

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

66

Thus using (H4), we have for each (x, t) ∈ D Z tZ |u(x, t)| ≤ |gv0 (x, t)| + G(x, t; y, s) q(y, s)Ψ (|u(y, s)|) dyds) 0 Ω Z tZ ≤ |gv0 |0 + G(x, t; y, s) q(y, s)dydsΨ (|u|0 ) 0

Ω

≤ |gv0 |0 + K0 |q|0 Ψ (|u|0 ) .

Hence |u|0 ≤ |gv0 |0 + K0 |q|0 Ψ (|u|0 ) .

(16)

Suppose, now that there exists u ∈ ∂U and λ ∈ (0, 1) such that u ∈ λ
(17)

u(x, t) = 0, (x, t) ∈ Γ,

(18)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

67

u(x, 0) = v (x) ,

x ∈ Ω.

(19)

We have seen that any solution u ∈ X of (17), (18), (19) is a solution of the operator equation zu = gv + Gη(u), where gv is given by (12) and Gη(u) is given by Z tZ G(x, t; y, s) η(u (y, s))dyds, Gγ(u)(x, t) = 0

Ω

(x, t) ∈ D.

We can show that z is continuous and completely continuous. Next, we proceed as in the proof Theorem 5.2 to show that any solution to u = λz (u) , for λ ∈ (0, 1) satisfies |u|0 6= M0 . Now, we apply Theorem 3.7 in27 to show that z has a fixed point u ∈ X. Then u (x, t; v0 ), where v0 is the fixed point of the operator Υ, is a solution of our problem.

5.4. Method of lower and upper solutions In this part we consider the case of an L2 − Carath´eodory multifunction of the form F (x, t, u) = [θ (x, t, u) , κ (x, t, u)] where θ, κ : D × R → R are such that θ (·, ·, u) , κ (·, ·, u) are measurable and θ (x, t, ·) is lsc and κ (x, t, ·) is usc. Thus F is an usc multifunction with nonempty, compact and convex values. u ∈ X is a solution of (1), (2), (3) if there exists f ∈ Lip (D) such that θ (x, t, u) ≤ f (x, t) ≤ κ (x, t, u) and (4), (5), (6) hold. Definition 5.2. z ∈ X is a lower solution of (1), (2), (3) if Dt z + Lz ≤ θ (x, t, z) , u(x, t) ≤ 0, z(x, 0) +

m X i=1

(x, t) ∈ D, (x, t) ∈ Γ,

βi (x)z(x, ti ) ≤ φ (x) ,

x ∈ Ω.

Z ∈ X is an upper solution of (1), (2), (3) if Dt Z + LZ ≥ κ (x, t, z) , (x, t) ∈ D, and the last two inequalities are reversed when we substitute Z for z.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

68

Theorem 5.5. Assume that c(x, t) ≥ c0 > 0 on D, βi (x) R ≤ 0 on Ω with P Pm G(x, ti ; y, 0)dy} −1 ≤ m β (x) ≤ 0 on Ω, and max { |β (x)| i x∈Ω i i=1 i=1 Ω < 1. Suppose that (1), (2), (3) has a lower solution z and an upper solution Z such that z (x, t) ≤ Z (x, t) . Then the problem has at least one solution u ∈ [z, Z]; i.e. z (x, t) ≤ u(x, t) ≤ Z (x, t) for every (x, t) ∈ D. Proof. Let δ (u) = max(z, min(u, Z)). Then δ : X → [z, Z] is continuous and bounded with |δ (u)|0 ≤ max(|z|0 , |Z|0 ). Consider the modified multifunction F1 : D × R → P (R) defined by F1 (x, t, u) = F (x, t, δ (u)). Then F1 is an L2 − Carath´eodory multifunction with nonempty, compact and convex values. Moreover, there exists ωF1 ∈ L2 (D) such that |F1 (x, t, u)| ≤ ωF1 (x, t) , ∀(x, t) ∈ D, ∀u ∈ R. It follows from Lemma 4.2 that the multivalued operator γ ◦ F1 is of usc type. Here F1 is the Nemitsky operator of F1 and γ is defined in Lemma 4.1. For v ∈ C(Ω) we consider the modified problem Dt u + Lu ∈ F1 (x, t, u), (x, t) ∈ D, u(x, t) = 0,

(22)

(x, t) ∈ Γ,

(23)

u(x, 0) = v (x) , x ∈ Ω, whose solutions are given by Z tZ G(x, t; y, s) f (y, s) dyds, u(x, t) = gv (x, t) + 0

Ω

(24)

(x, t) ∈ D.

where gv is given by (12) and f (x, t) ∈ F1 (x, t, u (x, t)), (x, t) ∈ D. Thus |u|0 ≤ |gv |0 + |ϕ|L2 |ωF1 |L2 := r0 . Let Q := {u ∈ X; |u|0 ≤ r0 }. Then Q is a nonempty, bounded, closed and convex subset of X. It follows from the properties of γ◦F1 that (γ ◦ F1 ) (Q) is compact and (γ ◦ F1 ) (Q) ⊂ Q. By a theorem of Bohnenblust and Karlin (see Cor. 11.3 (e) in11 ) the

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

69

operator γ ◦ F1 has a fixed point u1 . Then u1 (x, t; v0 ), where v0 is the fixed point of the operator Υ, is a solution of (22), (23), (24). To complete the proof we must show that u1 ∈ [z, Z].We show that u1 (x, t) ≥ z(x, t) for every (x, t) ∈ D. Suppose, on the contrary, that there exists (ζ, τ ) ∈ D such that u1 (ζ, τ ) < z (ζ, τ ) . Letting w (x, t) = u1 (x, t)−z(x, t), we see that w (ζ, τ ) < 0. w achieves a negative minimum at some point (ζ0 , η0 ) ∈ D. By the strong maximum principle (see16 ) we have either (ζ0 , η0 ) ∈ Γ or (ζ0 , η0 ) = (ζ0 , 0) with ζ0 ∈ Ω. Since w (x, t) ≥ 0 for (x, t) ∈ Γ, we see that min w (x, t) = w (ζ0 , 0) < 0. D

We now proceed as in the proof of Lemma 3.1 to complete the proof. Similarly, we show that u1 (x, t) ≤ Z(x, t) for every (x, t) ∈ D. Now, for u ∈ [z, Z] we have that δ (u) = u. Hence F1 (x, t, u) and F (x, t, u) coincide. Consequently, u1 is a solution of the original problem. Acknowledgements: The first draft of this paper was written while the author was visiting the Division of Applied Mathematics at Brown University through a Grant from The Arab Fund For Economic And Social Development, Kuwait. He is grateful to both Institutions and Professor John Mallet-Paret for making the visit possible. Also, he wishes to thank King Fahd University of Petroleum and Minerals for its constant support. References 1. J. P. Aubin and A. Cellina, Differential Inclusions, Springer Verlag, Berlin, New york, 1984. 2. J. P. Aubin and H. Frankowska, Set-Valued Analysis, Birkhauser, Boston, 1990. 3. K. Balachandran and K. Uchiyama, Existence of solutions of nonlinear integrodifferential equations of Sobolev type with nonlocal conditions in Banach spaces, Proc. Indian Acad. Sci. (Math. Sci.), 110 (2000), 225-232 4. A. Bressan and G. Colombo, Extensions and selections of maps with decomposable values, Studia Math. 90 (1988), 69-86 5. L. Byszewski, Theorems about the existence and uniqueness of solutions of a semilinear evolution nonlocal Cauchy problem, J. Math. Anal. Appl. 162 (1991), 494-505. 6. T. Cardinali, A. Fiacca and N. S. Papageorgiou, Extremal solutions for nonlinear parabolic problems with discontinuities, Mh. Math. 124 (1997), 119-131. 7. S. Carl and S. Heikkil¨ a, On a parabolic boundary value problem with discontinuous nonlinearity, Nonl. Anal. 15 (1990), 1091-1095. 8. S. Carl, C. Grossmann and C. V. Pao, Existence and monotone iterations for parabolic differential inclusions, Comm. Appl. Nonl. Anal. 3 (1996), 1-24. 9. J. Chabrowski, On nonlocal problems for parabolic equations, Nagoya Math. J. 93 (1984), 109-131.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

70

10. H. Deguchi, On weak solutions of parabolic initial value problems with discontinuous nonlinearities, Nonl. Anal. 63 (2005), e1107-e1117. 11. K. Deimling, Multivalued Differential Equations, W. de Gruyter, Berlin, 1992. 12. K. Deng, Exponential decay of solutions of semilinear parabolic equations with nonlocal initial conditions, J.Math. Anal. Appl. 179 (1993), 630-637. 13. E. Feireisl, A note on uniqueness for parabolic problems with discontinuous nonlinearities, Nonl. Anal. 16 (1991), 1053-1056. 14. E. Feireisl and J. Norbury, Some existence, uniqueness and nonuniqueness theorems for solutions of parabolic equations with discontinuous nonlinearities, Proc. Royal Soc. Edinburgh 119 A (1991), 1-17. 15. B. A. Fleishman and T. J. Mahar, A step function model in chemical reactor theory: Multiplicity and stability of solutions, Nonl. Anal. 5 (1981), 645-654. 16. A. Friedman, Partial Differential Equations of Parabolic Type, Prentice-Hall, Englewood Cliffs, New Jersey, 1964. 17. M. Frigon, Application de la th´eorie de la transversalit´e topologique a ` des probl`emes non lin´eaires pour des ´equations differentielles ordinaires, Dissertationes Math. 296 (1990), 1-79. 18. J. Hofbauer and P. L. Simon, An existence theorem for parabolic equations on RN with discontinuous nonlinearity, Electron. J. Qual. Theory Diff. Eqns. 2001, No. 8, pp.1-9 19. S. Hu and N. S. Papageorgiou, Handbook of Multivalued Analysis, Vol. I: Theory, Kluwer, Dordrecht, The Netherlands, 2000. 20. D. Jackson, Existence and uniqueness of solutions to semilinear nonlocal parabolic equations, J. Math. Anal. Appl. 172 (1993), 256-265. 21. A. A. Kerefov, Nonlocal boundary value problems for parabolic equations, Differential Equations 15 (1979), 52-54. 22. O. A. Ladyzhenskaya, V. A. Solonnikov and N. N. Uraltseva, Linear and Quasilinear Equations of Parabolic Type, Nauka, Moscow, 1967 [English translation: Amer. Math. Soc., Providence, R.I. 1968.] 23. A. Lasota and Z. Opial, An application of the Kakutani-Ky Fan theorem in the theory of ordinary differential equations, Bull. Acad. Pol. Sci. Ser. Sci. Math. Astronom. Phys. 13 (1965), 781-786. 24. G. M. Lieberman, Second Order Parabolic Differential Equations, World Scientific, River Edge, New York, 1996. 25. G. M. Lieberman, Nonlocal problems for quasilinear parabolic equations, in ”Nonlinear Problems in Mathematical Physics and related Topics I”, Kluwer 2002, 247-270. 26. Y. Lin, Analytical and numerical solutions for a class of nonlocal nonlinear parabolic differential equations, SIAM J. Math. Anal. 25 (1994), 1577-1594. 27. D. O’Regan, Integral inclusions of upper semi-continuous or lower semicontinuous type, Proc. Amer. Math. Soc. 124 (1996), 2391-2399 28. C. V. Pao, Nonlinear Parabolic and Elliptic Equations, Plenum Press, New York, 1992. 29. N. S. Papageorgiou, On the existence of solutions for nonlinear parabolic problems with nonmonotone discontinuities, J. Math. Anal. Appl. 205 (1997), 434-453.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

71

30. V. N. Pavlenko and O. V. Ul’yanova, Method of upper and lower solutions for parabolic-type equations with discontinuous nonlinearities, Differential Equations 38 (2002), 520-527. 31. R. Pisani, Problemi al contorno per operatori parabolici con non linearita discontinua, Rend. Ist. Mat. Univ. Trieste 14 (1982), 85-98. 32. T. Pruszko, Topological degree methods in multivalued boundary value problems, Nonl. Anal. 5 (1981), 953-973 33. J. Rauch, Discontinuous semilinear differential equations and multiple valued maps, Proc. Amer. Math. Soc. 64 (1977), 277-282. 34. A. A. Samarskii, Some problems in differential equation theory, Differential Equations 16 (1980), 1221-1228. 35. P. N. Vabishchevich, Nonlocal parabolic problems and the inverse heatconduction problem, Differential Equations 17 (1981), 761-765.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

72

Periodic solutions of nonlinear parabolic equations with measure data and polynomial growth in |∇u| Abderrahmane El Hachimi∗ and Abdelilah Lamrani Alaoui† UFR Math´ ematiques Appliqu´ ees et Industrielles, Facult´ e des Sciences B.P 20, El Jadida, Morocco E-mails: ∗ [email protected], † [email protected] In this paper we are interested in the problem:  ∂u  − ∆p u = f (x, t, u, ∇u) + λµ in QT  ∂t (P) u=0 on ΣT   u(0) = u(T ) in Ω

where f is a Carath´eodory function with polynomial growth in |∇u|, λ is a real number, µ a bounded Radon measure and p ≥ 2.

Keywords: Periodic solution; Nonlinear parabolic equation.

1. Introduction Periodic solutions of parabolic problems were studied by many authors and in many context particulary when the elliptic operator is linear (see10,12 and the references therein). In the case of measure datum the unique paper we found is1 where (P) with p = 2 was studied. In that work the authors firstly give necessary conditions on the datum for existence, they show that a size and regularity information is consequence of existence in some cases. Secondly they give existence in the case of special measure, that is an L1 function. In the present work we generalize in different directions the results in,1 firstly the elliptic operator is no more linear it is the so called degenerate p−Laplacian, that is −∆p u = −div(|∇u|p−2 ∇u), p ≥ 2. Secondly the existence in the natural growth case is given for sufficiently regular measures (in some capacity sense) but not necessarily in L1 (QT ). The notion of capacity considered here is the one adapted to the parabolic

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

73

p−Laplacian operator due to Droniou et al.8 Let us mention that the absence of information about the initial value of the solution causes some technical difficulties. Throughout the paper we adopt the following notations: • Ω an open bounded subset of IRN , ∂Ω its boundary, T > 0, QT = Ω × (0, T ) and ΣT = ∂Ω × (0, T ), 0 0 0 • V 0 = Lp (0, T ; W01,p(Ω)) and V 0 = Lp (0, T ; W −1,p (Ω)) its dual where p p0 = , p− 1 0 ∂u ∈ V0 + L1 (QT ) • W = u ∈ V0 such that ∂t • Wr1,1 (QT ) is the set of all measurable functions u such that Z ∂u r r r |u| + |Du| + | | dxdt < +∞ ∂t QT • C b (Ω) the space of bounded and continuous function on Ω, • Mb (QT ) the space of bounded Radon measures, • M0 (QT ) the subspace of Mb (QT ) which don’t charge sets of parabolic capacity zero (see section 3), 0 • ., . is the duality between V 0 and V 0 . 0 • < ., . > is the duality between W −1,p (Ω) and W01,p (Ω), • |.| is the Lebesgue measure of a borelian set, the euclidian norm of a vector or the absolute value of a real number, • k.kX the norm of an element of a Banach space X. If X = Lp with p ∈ [1, +∞] it will be noted k.kp , • C is a generic positive constant where the dependence on parameters will be indicated only if needed. The following hypothesis is common to all sections. f is a Carath´eodory function: measurable in (t, x)for all(r, ξ) ∈ IRN +1

(1)

and continuous in (r, ξ) for almost every (t, x) ∈ QT . The paper is organized as follows. Section 2 is devoted to necessary conditions of existence, regularity of the measure and nonexistence result are established. Subsequently, we present existence result in the case of natural growth and some regular measures datum. The paper is organized as follows. Section 2 is devoted to necessary conditions of existence, regularity of the measure and nonexistence result

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

74

are established. Subsequently, we present existence result in the case of natural growth and some regular measures datum. 2. Necessary conditions for existence In this section we are interested in the following problem: for λ ∈ IR find u such that: u ∈ Lp (0, T, W01,p (Ω)) ∩ C(]0, T [, L1(Ω)) f (x, t, u, ∇u) ∈ L1loc (QT ), 0 ∂u − ∆p u ≥ f (x, t, u, ∇u) + λµ in D (QT ) ∂t

(2)

u(0) = u(T ) in Mb (Ω), where µ is nonnegative bounded Radon measure.

(3)

The periodicity condition u(0) = u(T ) in Mb (Ω) is understood in the following sense Z lim+ (u(x, t) − u(x, T − t))ϕdx = 0 t→0

Ω

for all ϕ ∈ C b (Ω).

2.1. Nonexistence in supercritical case (q > 2(p − 1)) In this subsection we prove that solutions to (2) does not exists for all values of λ if the growth of f in |∇u| is sufficiently large. We assume that there exists ]ε, σ[ open in ]0, T [,

(4)

q > 2(p − 1)

(5)

and a constant C0 > 0 such thatf (x, t, s, ξ) ≥ C0 |ξ|q for almost every (t, x) ∈ ]ε, σ[ ×Ω := Qε,σ and for all (s, ξ) ∈ IRN +1 (6)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

75

and finally µ(]ε, σ[×Ω) > 0.

(7)

The main result of this subsection is the following: Theorem 2.1. Assume that (1), (3) and (4)-(7) hold, then there exists finite λ∗ such that (2) does not have any solution for λ > λ∗ . Remark 2.1. The condition (5) is reduced to q > 2 in the case p = 2 which is the superquadratic case treated in1 and.2 Now we prove Theorem 2.1. Proof. We follow here the proofs of 1 and2 with some changes due to the nonlinear feature of the problem (2). If u is a solution of (2), then by (6) we have 0 ∂u − ∆p u ≥ C0 |∇u|q + λµ in D (Qε,σ ) ∂t let φ ∈ C ∞ 0 (]0, T [×Ω) such that φ ≥ 0 and φ(ε) = φ(σ) = 0. Multiply (2) by φ and integrate from ε to σ we obtain Z Z Z Z λ φdµ ≤ |∇u|p−1 |∇φ| − C0 |∇u|q φ −

Qε,σ

Qε,σ

Qε,σ

(8)

u∂t φ.

(9)

Qε,σ

This inequality can be extended to φ ∈ C 1 ([0, T ]; L∞ (Ω)) ∩ L∞ (0, T ; W01,∞(Ω)), φ ≥ 0 and φ(ε) = φ(σ) = 0. But φt = −∆G(φt ) = −∆(Gφ)t where G is the Green Kernel on Ω. Then from (9) and using twice Young inequality we obtain ! q Z Z Z 0 |∇(Gφ)t |q |∇φ| q−p+1 + . (10) φdµ ≤ C λ 1 p−1 Qε,σ Qε,σ Qε,σ φ q−1 φ q−p+1 For φ = (σ−t)q (t−ε)q Φ, where Φ is a solution of the eigenvalue problem −∆φ = λ1 φ in Ω, φ > 0 in Ω, φ = 0 on ∂Ω.

(11)

Where λ1 is the first eigenvalue of −∆ in Ω. We have q (Gφ)t = Φ(t − ε)q−1 (σ − t)q−1 (σ + ε − 2t) λ1 1 Using (5) and the fact that φ ∈ W01,∞ (Ω) and α ∈ L1 (Ω) for all α < 1. φ the two terms of the right hand side of (10) are finite. Finally using (3) the existence of a finite λ∗ such that: existence for (2) implies λ ≤ λ∗ .

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

76

2.2. Existence implies regularity of the measure We recall first that a compact set K in QT is of Wr1,1 (QT )-capacity zero where r ≥ 1, if there exists a sequence of C0∞ (QT ) functions (φn )n such that : 0 ≤ φn ≤ 1 a.e in QT , φn ≡ 1 on K, φn → 0 in Wr1,1 (QT ) and a.e. in QT . (12) The main result of this subsection concerns the problem (2) when: There exists q > p − 1,

(13)

and there exists C1 and C2 such that f (x, t, s, ξ) ≥ C1 |ξ|q − C2 in QT × IRN +1 . (14) Theorem 2.2. Assume that (1), (3) hold, and that f satisfies (14) and (13). If (2) has a solution then µ does not charge the sets of q Wr1,1 (QT )-capacity zero where r = . q−p+1 Remark 2.2. The above statement means that: K compact, Wr1,1 (QT )-capacity(K) = 0 ⇒ µ(K) = 0. This result extend to the p−Laplacian the one of 1 in the Laplacian case q . (p = 2) where r was q−1 In the case q > 2(p − 1) both size and regularity conditions are necessary for existence. Proof. From (2) and (14) we have: 0 ∂u − ∆p u ≥ C1 |∇u|q − C2 + λµ in D (QT ). ∂t

(15)

Let K a compact set of Wr1,1 (QT )-capacity zero and (φn )n ⊂ C ∞ 0 (QT ) such that (12) is satisfied. Multiplying (15) by ψn = φn r (where r is given in Theorem 2.2) we obtain Z Z Z Z ∂ψn q p−1 λµ(K) + C1 |∇u| ψn ≤ − u + |∇u| |∇ψn | + C2 ψn . ∂t QT QT QT QT (16)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

77

By Young inequality we have: Z Z p−1 |∇u| |∇ψn | ≤ ε QT

q

QT

|∇u| ψn + C(ε)

Z

QT

|∇φn |r .

For ε sufficiently small and with (17) into (16) we obtain Z Z Z |∇φn |r + C2 u.φn r−1 ∂t φn + C λµ(K) ≤ −r QT

QT

QT

φrn .

(17)

(18)

Passing to the limit using Vitali theorem and (12) we have µ(K) = 0 3. Existence in the case of natural growth and regular measure data In this section we focus our attention on the problem (Proof .) where f takes the form f (x, t, s, ξ) = −g(s)|ξ|p with g a continuous fonction on IR such that g ∈ L1+ (IR). That is : ∂u − ∆p u + g(u)|∇u|p = µ in QT u = 0 on ΣT u(0) = u(T ) in Ω. ∂t

(19)

(20)

Remark 3.1. Because p ≥ 2 we have p ≤ 2(p − 1) then no size condition on the measure is needed in this section and the parameter λ is dropped. The lack of initial data by the periodicity condition leads to some difficulties which we overcome by considering the following problem.   u ≥ g2 a.e. in QT     ∂u p − ∆p u + g(u)|∇u| = µ in QT (21) ∂t   u=0 on ΣT    u(0) = u(T ) in Ω 0

where µ is a regular measure in the sense µ ∈ L(QT ) = V0 + L1 (QT ) + ∂t (V0 ∩ L∞ (QT )) that is µ = g 0 + g 1 + ∂ t g2 0

where (g0 , g1 , g2 ) ∈ V0 × L1 (QT ) × (V0 ∩ L∞ (QT )). Before giving the meaning of solutions for the problem (21) let us review some well known results we use here and explain the reason for that choice of measures.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

78

3.1. Some known results We begin with a decomposition for measures which do not charge sets of parabolic capacity zero. First, recall that, if U ⊂ QT is an open set, parabolic capacity of U is defined as follows: Capp (U ) = inf {kukW : u ∈ W, u ≥ χu a.e in QT } Then for any borelian set B ⊂ QT Capp (B) = inf {Capp (U ), U open subset of QT , B ⊂ U } . For more properties and details on the parabolic capacity see.8 Proposition 3.1. (theorem 2.28 8 ) If µ ∈ M0 (QT ), the subspace of measures in Mb (QT ) which don’t charge sets of parabolic capacity zero. 0 Then there exists (g0 , g1 , g2 ) ∈ V0 × L1 (QT ) × V0 such that : Z T Z T Z Z φdµ = g1 φdxdt− < g0 , φ > dt+ < (φ)t , g2 > dt ∀φ ∈ D(QT ). QT

0

QT

0

We now give a trace and integration by part results which make it possible to define and treat periodicity in (21). Proposition 3.2. (theorem 1.1 13 and Lemma 2 15 ) We have: W ,→ C([0, T ]; L1(Ω)), and for v ∈ W, φ Lipschitz bounded real function from IR to IR, with φ(0) = 0 we have for all t ∈ [0, T ] Z Z v(x,t) Z Z v(x,0) Z t ∂v(σ) , φ(v(σ)) > dσ = dx φ(σ)dσ − dx φ(σ)dσ < ∂t Ω 0 Ω 0 0 Moreover if v(0) = v(T ) in L1 (Ω) we have

∂ , φ(v) = 0. ∂vt

Now we are ready to give the main result of this section. 3.2. Definition and main result (,45 and7 ) Let µ be a measure in L(QT ) with a decomposition (g0 , g1 , g2 ), a measurable function u ≥ g2 is said to be renormalized solution of (21) if Tk (u) ∈ V0 ,

(22)

Tk (u − g2 ) ∈ V0 ,

(23)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

79

for every k > 0, and

Z

lim

n→+∞ {n≤u−g2 ≤n+1}

|∇u|p dxdt = 0,

(24)

and, for every S ∈ W 2,∞ (IR) such that S 0 has compact support in IR+ , S(0) = 0, ∂t S(u − g2 ) − div(S 0 (u − g2 )|∇u|p−2 ∇u) 00 +S (u − g2 )|∇u|p−2 ∇u.∇(u − g2 ) 0 +g(u)|∇u|p S (u − g2 ) 0

00

0

= S (u − g2 )g1 + G0 S (u − g2 )∇(u − g2 ) − div(G0 S (u − g2 ))

(25)

in the sense of distributions, with g0 = div(G0 ), S(u − g2 )(0) = S(u − g2 )(T ).

(26)

We state our main result Theorem 3.1. Let µ a measure in L(QT ) and assume that (19) holds, then (21) has a solution in the sense of definition above. The proof of this result is divided into three classical steps: - Regularization. - Existence of approximate solutions. - A priori estimates and passing to the limit. 3.3. Regularization and existence of approximate solutions Let (g0 , g1 , g2 ) be a decomposition of µ, we regularize (21) by the sequence of problems − ∆p un + g(un )|∇un |p − nTn (un − g2+ )− = µn in QT un = 0 on ΣT un (0) = un (T ) in Ω ∂un ∂t

(27)

where, µn = g0n + g1n + (g2n )t , with gin ∈ Cc∞ (QT ), i = 1, 2, 3 and kµn kL1 (QT ) ≤ C, (28) and 0

g0n → g0 in V0 ,

(29)

g1n → g1 in L1 (QT ),

(30)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

80

g2n → g2 in V0 and a.e inQT , with kg2n k∞ ≤ kg2 k∞ .

(31)

Let us prove the existence of approximate solutions, to do this we show for fixed n ∈ IN existence of ordered lower and upper solutions to (27). We begin by showing existence of an upper solution. Let β˜n ∈ W01,p (Ω) ∩ C 1 (Ω) be the solution of the Torsion problem: −∆p u = kµn k∞ + n2 in Ω, u = 0 on ∂Ω,

(32)

−∆p u + g(u)|∇u|p = −kµn k∞ in Ω, u = 0 on ∂Ω,

(33)

and αn ∈ W01,p (Ω) ∩ L∞ (Ω) , see,6 the solution of the problem

By taking βn = β˜n + β where β is a large real number such that βn ≥ max(αn , g2 ) almost everywhere in QT . 3.4. A priori estimates and passage to the limit We have constructed a pair of ordered lower and upper solutions to (27), now we use9 to derive the existence of a solution un ∈ Q0 ∩ L∞ (QT ), un ∈ Q([0, T ]; Ls (Ω)) for all s ≥ 1, and αn ≤ un ≤ βn a.e. in QT We collect some a priori estimates, we begin by Lemma 3.1. (nTn (un − g2n )− )n is bounded in L1 (QT ). Proof. Fix k > kg2 k∞ , let h R> 0 and consider ϕn = Th (un − s Tk (un ))exp(−G(un )) where G(s) = 0 g(σ)dσ for all s in IR as a test function in (27), we obtain Z Z Z |∇un |p exp(−G(un ))χ{k<|un |
QT

QT

but exp(−kgkL1(IR) ) ≤ exp(−G(s)) ≤ exp(kgkL1 (IR) ) for all s ∈ IR

(34)

then (28) gives Z −n Tn (un − g2n )− Th (un − Tk (un ))exp(−G(un )) ≤ Ch, QT

the choice of k gives by dividing on h and passing to the limit as h tends to 0 Z Tn (un − g2n )− ≤ C. n QT

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

81

The proof is then complete. Proposition 3.3. For fixed k > 0, let un be solution to (27), then for all n ∈ IN Z |∇Tk (un )|p ≤ Ck, (35) QT

Z

|∇Tk (un − g2n )|p ≤ C(k + 1),

(36)

lim sup |{|un − g2n | ≥ h}| = 0.

(37)

QT

h→+∞ n∈IN

and



Z

 lim sup

h→+∞

n



h≤un −g2n ≤h+k

 |∇un |p dxdt = 0.

(38)

Moreover there exists a measurable function u : QT → IR such that u ≥ g2 a.e. in QT and Tk (u), Tk (u − g2 ) belong to V0 and, up to a subsequence

un → u a.e. in QT , Tk (un − g2n ) * Tk (u − g2 ) in V0 and a.e. in QT . (39)

Finally lim

Z

h→+∞ h≤u−g2 ≤h+k

|∇u|p dxdt = 0.

(40)

Proof. In all that follows, we set vn = un − g2n and v = u − g2 . By taking Tk (un )exp(−G(un )) as test function in (27) and using the periodicity of un , (34) we obtain R R C QT |∇Tk (un )|p ≤ QT µn Tk (un )exp(−G(un )) R (41) +n QT Tn (un − g2n )− Tk (un )exp(−G(un )). Then (28) and Lemma 3.1 gives (35). For (36) we have Z Z p |∇Tk (vn )| = |∇Tk (TMk (un ) − g2n ) |p QT ZQ T |∇ (TMk (un ) − g2n ) |p χ{|vn |≤k} = QZT ≤C (|∇TMk (un )|p + |∇g2n |p ) χ{|vn |≤k} QT

≤ C(Mk + 1) ≤ C(k + 1)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

82

where Mk = kg2 k∞ + k. Now we prove (37), to do this we follow.3 Fix h > 0, let ε ∈ ]0, h[ we have : |{|vn | ≥ ε}| = |{|Th (vn )| ≥ ε}| ≤ Then |{|vn | ≥ ε}| ≤ C taking for example ε =

Z

QT

kTh (vn )kpp |Th (vn )|p = . p ε εp

k∇Th (vn )kpp h+1 ≤C p , εp ε

h we obtain: 2 h h+1 | |vn | ≥ |≤C p 2 h

but p ≥ 2, then (37) is proved. Now we show the existence of a measurable function u such that un → u a.e in QT , and Tk (u), Tk (u − g2 ) ∈ V0 for all k ∈ IN.

We adapt to our case the method used in8 (see also3 ). Given ε > 0, fix (by the help of (37)) k ∈ IN such that k | < ε, independently of n. | |vn | > 2

(43)

Let Tk ∈ C 2 (IR) nondecreasing function such that Tk (s) = s for |s| ≤ Tk (s) = sign(s)k for |s| > k. 0 Taking in (27) the test function Tk (vn )ϕ where ϕ ∈ Cc∞ (QT ) we get 0

(42)

k 2

and

00

(Tk (vn ))t − div(Tk (vn )|∇un |p−2 ∇un ) + |∇un |p−2 ∇un .∇vn Tk (vn ) 0 0 +g(un )|∇un |p Tk (vn ) − nTn (vn− )Tk (vn ) 0

0

0

00

= Tk (vn )g1n − div(Gn0 Tk (vn )) + Tk (vn )Gn0 ∇vn , div(Gn0 ) = g0n .

(44)

Tk has a compact support, thanks to (31) and Lemma 3.1 the quan00 0 00 tities |∇un |p−2 ∇un .∇vn Tk (vn ), nTn (vn− )Tk (vn ) and Tk (vn )Gn1 ∇vn are 0 0 bounded in L1 (QT ), so are g(un )|∇un |p Tk (vn ) and Tk (vn )g1n because of 0 0 (19) and (30). Similarly, Gn0 Tk (vn ) and Tk (vn )|∇un |p−2 ∇un are bounded 0 0 in (Lp (QT ))N , then from (44) (Tk (vn ))t is bounded in V0 + L1 (QT ). Since Tk (vn ) is bounded in V0 a compactness result from17 leads to compactness of (Tk (vn ))n in L1 (QT ). Thus for a subsequence, it also a Cauchy sequence in measure. Take τ > 0, for n and m large we have |{|Tk (vn ) − Tk (vm )| > τ }| ≤ ε.

(45)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

83

By the choice of Tk :

k k |{|vn −vm | > τ }| ≤ | |vn | > |+| |vm | > |+|{|Tk (vn )−Tk (vm )| > τ }|. 2 2 (46) Then from (43): |{|vn − vm | > τ }| ≤ 3ε,

(47)

so that vn = un − g2n is a Cauchy sequence in measure. Passing to subsequence and using (31) there exists a measurable function u such that un converges to u almost everywhere and Tk (un − g2n ) * Tk (u − g2 ) in V0 , and Tk (un ) * Tk (u) in V0 .

Moreover, Lemma 3.1 and Fatou’s lemma give v − = 0 a.e. in QT then u ≥ g2 a.e. in QT . Now we prove (40), it suffices to prove (38). Take ψ(vn ) = Tk+ (vn − Th (vn )) as test function in (27) we obtain Z ∂vn |∇un |p−2 ∇un ∇vn χ{h≤vn ≤h+k} , ψ(vn ) + ∂t QT Z Z Z + g(un )|∇un |p ψ(vn )−n Tn (vn− )ψ(vn ) = g1n ψ(vn ) QT QT R T QT + 0 hg0n , ψ(vn )i

the truncation function satisfies the sign condition and ψ is positive then Z Z Tn (vn− )ψ(vn ) = 0, g(un )|∇un |p ψ(vn ) ≥ 0 and n QT

QT

and due to the integration result in Proposition 3.2 we have R R p−2 n |∇un |p χ{h≤v QT Z Zn ≤h+k} − QT |∇un | ∇un ∇g2 χ{h≤vn ≤h+k} ≤ |g1n | + |Gn0 ||∇ψ(vn )| {vn ≥h}

then C

Z

{h≤vn ≤h+k}

(48)

QT

|∇un |p ≤

Z

{h≤vn ≤h+k}

0

|∇Gn0 |p +

Z

{vn ≥h}

|g1n |+

Z

|∇g2n |p

{h≤vn ≤h+k} 0

by (37) and the equi-integrability of the sequences (g1n )n , (|Gn0 |p )n and (|∇g2n |p )n in L1 (QT ) we obtain (38), The boundedness of (Th+k (vn ))n in V0 and (38) gives (40).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

84

Return to the proof of Theorem 3.1. The purpose in what follows is the strong convergence Tk (vn ) → Tk (v) in V0

(49)

to this end we use almost the same techniques as in8 and.13 Recall first the following time regularization of Tk (v) due to R. Landes in the stationary case and adapted to the parabolic one in13 and.8 Let (zν )ν be a sequence of positive function such that: zν ∈ W01,p (Ω) ∩ L∞ (Ω), kzν k∞ ≤ k, ∀ν > 0, zν → Tk (v)(T ) a.e. in Ω as ν tends to infinity, 1 lim kzν kW 1,p (Ω) = 0, 0 ν→+∞ ν then note Tk (v)ν the unique solution of the problem:   ∂Tk (v)ν = ν(Tk (v) − Tk (v)ν ), ∂t  T (v) k ν (0) = zν

which has the form

Tk (v)ν (t) = Then we have (see11 )

Z

t

0

νexp(ν(s − t))v(s)ds + zν exp(−νt)

(50)

Tk (v)ν → Tk (v) in V0 and a.e. in QT . kTk (v)ν k∞ ≤ k, ∀ν > 0. We are ready to prove (49). Let (un )n be a sequence of solutions of (27) where µn satisfies (28), and let u given by Proposition 3.3. For h > 2k, we introduce as in13 and8 the following function wn = T2k (vn − Th (vn ) + Tk (vn ) − Tk (v)ν ) we will note ε(n, ν, h) all the quantities (possibly different) such that lim

lim

lim ε, ν, h) = 0

h→+∞ ν→+∞ n→+∞

and this will be the order in which the parameters we use will tends to infinity, that is, first n, then ν and finally h. Choosing wn as test function in (27) we have Z Z ∂vn p−2 , wn + |∇un | ∇un ∇wn + g(un )|∇un |p wn QT Z QT Z∂t g1n wn + g0n , wn Tn (vn− )wn = −n QT

QT

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

85

remark that wn = T2k (TM (vn ) − Th (vn ) + Tk (vn ) − Tk (v)ν ) where M = h + 2k, then (wn )n is bounded in V0 and wn * T2k (v − Th (v) + Tk (v) − Tk (v)ν ) in V0 and a.e. in QT and lim

n→+∞

Z

g1n wn

QT lim g0n , wn n→+∞

=

Z

QT

g1 T2k (v − Th (v) + Tk (v) − Tk (v)ν ),

= g0 , T2k (v − Th (v) + Tk (v) − Tk (v)ν )

the fact that Tk (v)ν → Tk (v) in V0 and a.e. in QT gives Z Z g1 T2k (v − Th (v)), g1 T2k (v − Th (v) + Tk (v) − Tk (v)ν ) = lim ν→+∞

QT

QT

lim g0 , T2k (v − Th (v) + Tk (v) − Tk (v)ν ) = g0 , T2k (v − Th (v))

ν→+∞

by the Lebesgue theorem lim

h→+∞

Z

QT

g1 T2k (v − Th (v)) = 0.

Moreover

g0 , T2k (v−Th (v)) =

Z

{h≤v≤h+2k}



 G0 ∇v ≤ kG0 kp0 

Z

{h≤v≤h+2k}

 p1

 |∇v|p  ,

then, because of (40) we have Z Z ∂vn p−2 Tn (vn− )wn = ε(n, ν, h) |∇un | ∇un ∇wn − n , wn + ∂t QT QT but n

Z

QZ T

Tn (vn− )wn

=n =n

=n

ZQ T ZQ T

QT

Tn (vn− )T2k (vn+ − vn− − Th (vn+ − vn− ) + Tk (vn+ − vn− ) − Tk (v)ν ) Tn (vn− )T2k (−vn− − Th (−vn− ) + Tk (−vn− ) − Tk (v)ν ) Tn (vn− )T2k (Th (−vn− ) − vn− − (Tk (vn− ) + Tk (v)ν ))

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

86

The positivity of v and (50) gives Tk (v)ν ≥ 0 thenR Tk (vn− ) + Tk (v)ν ≥ 0, and clearly Th (−vn− ) − vn− ≤ 0 a.e. in QT , then n QT Tn (vn− )wn ≤ 0 and finally Z ∂vn , wn + |∇un |p−2 ∇un ∇wn ≤ ε(n, ν, h) ∂t QT the rest of the proof is the same as13 as far as the term

∂vn , wn is ∂t

concerned and in8 for the others. Periodicity condition passes to the limit in the following way: By (44) one has 0

∂t S(un − g2n ) converge strongly in V0 + L1 (QT ), but S(un − g2n ) converge strongly in V0 . Then by Lemma 3.2 S(un − g2n ) → S(u − g2 ) strongly in C([0, T ]; L1 (Ω)). But S(un − g2n )(0) = S(un − g2n )(T ) then S(u − g2 )(0) = S(u − g2 )(T ) in Ω. And periodicity is now proved.

References 1. N. Alaa, M. Iguernane, Weak periodic solutions of some quasilinear parabolic equations with data measures , Journal of Inequalities in Pure and Applied Mathematics. 3 (3),No. 46, (2002). 2. N. Alaa, M. Pierre, Weak solutions of some quasilinear elliptic equations with data measures, SIAM J. Math. Anal. 24 (1993), n. 1, 23-35. 3. P. Benilan, L. Boccardo, T. Gallou¨et, R. Gariepy, M. Pierre, J. L. V` azquez, An L1 theory of existence and uniqueness of nonlinear elliptic equations, Ann. Scuola Norm. Sup. pisa CI. Sci., 22, (1995),641-655. 4. D. Blanchard, A. Porretta, Nonlinear parabolic equations with natural growth terms and measure initial data. Annali della Scuola Normale Superiore di Pisa - Classe di Scienze S´er. 4, 30 . 3-4 (2001), p. 583-622 5. D. Blanchard, F. Murat, Renormalised solutions of nonlinear parabolic problems with L1-data: existence and uniqueness, Proc. Royal Soc. Edinburgh Sect. A 127 (1997), 1137-1152.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

87

6. L. Boccardo, S. Segura, C. Trombetti, Bounded and unbounded solutions for a class of quasi-linear elliptic problems with a quadratic gradient term, J. Math. Pures Appl., (9) 80 (2001), 919–940. 7. G. Dal Maso, F. Murat, L. Orsina, A. Prignet,Definition and existence of renormalized solutions of elliptic equations with general measure data, C. R. Acad. Sci. Paris S´er. I Math. 325 (1997), 481-486. 8. J. Droniou, A. Porretta, A. Prignet, Parabolic capacity and soft measures for nonlinear equations, Potential Analysis, Volume 19, Number 2, September 2003, pp. 99-161(63) 9. A. El Hachimi And A. Lamrani Alaoui, Existence of stable periodic solutions for quasilinear parabolic problems in the presence of well-ordered lower and upper-solutions,proceeding of EJDE, 9 (2002), pp. 1-10. 10. P. Hess. periodic-parabolic boundary value problem and positivity, Pitman Res. Notes Mathematics, Vol. 247, Longman Sci. Tech., Harlow, 1991. 11. R. Landes, On the existence of weak solutions for quasilinear parabolic boundary value problems, Proc. Roy. Soc. Edinburgh Sect. A 89 (1981), 217-237. 12. P. Polacik, Parabolic equations: asymptotic behaviour and dynamics on invariant manifolds, In: Handbook of Dyn. Syst., Vol. 2, North-Holland, Amsterdam, pp. 835–883. 13. A. Porretta, Existence results for nonlinear parabolic equations via strong convergence of truncations , Ann. Mat. Pura ed Applicata (IV) 177 (1999), pp. 143–172. 14. A. Porretta; Nonlinear equations with natural growth terms and measures data, proceeding of EJDE, 9 (2002), pp. 183-202. 15. J. M. Rakotoson, A compactness lemma for quasilinear problems: Application to parabolic equations, J. Funct. Anal., Vol. 106, No. 2, pp. 1163-1175, 1992. 16. S. Segura de Leon, Existence and uniqueness for L1 data of some elliptic equations with natural growth, Advances in Diff. Eq., to appear. 17. J. Simon, compact sets in Lp (0, T ; B), Ann. Math. Pura Appl. , 146 (4) (1987), 65-96.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

88

Existence and L∞ -regularity results for some nonlinear elliptic Dirichlet problems Ahmed Youssfi Department of Mathematics and Informatic Faculty of Sciences Dhar El Mahraz B.P 1796 Atlas, F` es, Morocco E-mail: [email protected] We summarize recent lectures devoted to the study of the existence and L ∞ regularity results of some Dirichlet problems associated to equations having degenerated coercivity in the principal part. Keywords: Orlicz spaces; Nonlinear elliptic equations; Degenerate coercivity; A priori estimates; L∞ -estimates; Rearrangements.

1. Introduction During the 19th century, H¨ older regularity results to the Dirichlet problem for linear elliptic equations has been object of much research and has developed by many authors. Far from being complete, the story goes that motivated in solving the Hilbert’s 19th conjecture, E. DeGiorgi11 in 1957, the first, proved his famous regularity result which asserts that local weak solutions for the linear problem ( (ai,j uxi )xj = 0 weakly in Ω, (1) u∈W01,2 (Ω), where the coefficients x → ai,j (x), i, j = 1, ..., N are only bounded and measurable and satisfying the uniform ellipticity condition ai,j ξi ξj ≥ α|ξ|2 , a.e. in Ω, ∀ξ ∈ IRN , for some α > 0,

(2)

are H¨ older continuous by studying local pointwise estimates. The global bound appears in the 1960’s in the works of G. Stampacchia15,16 and Ladyzhenskaja-Ural’tseva.13 Following the methods of DeGiorgi, Ladyzhenskaja and Ural’tzeva established that weak solutions of quasilinear elliptic equations are H¨ older continuous. Since, several L∞ −regularity results to

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

89

the Dirichlet problem, linear or nonlinear, more general than (1) have been established. Let us consider the problem ( −div(a(x, u)∇u) = f in Ω, (3) u∈H01 (Ω) ∩ L∞ (Ω), with the degenerate coercivity a(x, s) ≥

α with α > 0 (1 + s)θ

where Ω is a bounded open subset of IRN , N ≥ 2, and f ∈ Lm (Ω) with m > N/2.

The existence of bounded solutions for (3) was proved in Alvino et al.,2 Boccardo-Brezis6 and in Boccardo et al.7 The result was then extended in Alvino et al.1 for the nonlinear elliptic problem ( −diva(x, u, ∇u) = f in Ω, (4) u∈W01,p (Ω) ∩ L∞ (Ω), p > 1 with degenerate coercivity a(x, s, ξ) · ξ ≥

α |ξ|p with α > 0 (1 + s)θ(p−1)

(5)

and where f ∈ Lm (Ω) with m > N/p.

(6)

When we replace in (4) f by −divg, where

|g| ∈ Lm (Ω) with m > N/(p − 1),

(7)

5

Benkirane-Youssfi have proved that problem (4) has bounded solutions. Strongly nonlinear elliptic problems are also treated. It is so the problem ( A(u) + H(x, u, ∇u) = f − divg in Ω, (8) u∈W01,p (Ω) ∩ L∞ (Ω), where A(u) = −diva(x, u, ∇u) has uniform coercivity and f and g satisfying (6) and (7) respectively, has at least one solution Boccardo et al.8 and Ferone et al.10 In Boccardo et al.9 the authors have considered the problem (8) with A(u) = −div(a(x, u)∇u) having degenerate coercivity and g = 0 and have shown that it has at least one solution in H01 (Ω) ∩ L∞ (Ω). The general case 1 < p < +∞ with g = 0 and a(x, s, ξ) satisfying (5) Trombetti21 established that (8) has at least one solution.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

90

Here we present a survey on some recent existence and L∞ -regularity results for the Dirichlet problem associated to nonlinear elliptic equations on bounded open subset Ω of IRN , N ≥ 2, involving the following noneverywhere nonlinear differential operator of Leray-Lions type A(u) := −div a(x, u, ∇u). The function a may have nonstandard growth and satisfies the condition: a(x, s, ξ)·ξ ≥ M

−1

(M (h(|s|)))M (|ξ|),

(9) p

where M is an N -function (take as example M (t) = t , p > 1) and h : IR+ → ]0, +∞[ is a continuous decreasing function such that : 0 < h(0)≤1 and its primitive Z s h(t)dt (10) H(s) = 0

1 is unbounded (take for instance h(t) = (1+t) θ , 0 ≤ θ ≤ 1.) p It is worth recalling that in the L setting, a priori bounds for solutions of nonlinear elliptic Dirichlet problems can be derived by using either the method of Stampacchia15 or by means of the rearrangements techniques widely developed in Talenti.17–20 Observe that because of the lack of coercivity generated by the assumption (9), the operator A degenerates when its second argument has large values. Therefore, the classical existence results in Leray-Lions14 and Gossez-Mustonen12 used to prove the existence of a solution for the problem (4) cannot be applied even if the datum f is very regular. To get rid of this difficulty, we will consider approximate equations in which we introduce a truncation, in order to get non degenerate approximated problems which thus have solution. Once this done, some a priori estimates on the solutions of these problems are proved. To complete the ingredients which allow us to pass to the limit, we prove the almost everywhere convergence of the gradients of solutions.

2. Standard growth One stands in the Lp setting. Let M (t) = tp with 1 < p < N. It is known that the Dirichlet problem A(u) = f in Ω, (11) u=0 on ∂Ω, has bounded solutions provided that the datum f satisfies (6). Even in the N case when h is the constant function, the limit case f ∈ L p yields a solution

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

91

u in a suitable Orlicz space of exponential type and it is (in general) not bounded. Precisely, H(u) ∈ Lφ (Ω) with φ(t) = exp(t) − t − 1 and H is the function defined in (10). N We consider a subclass of L p (Ω) larger than (6) which guaranties the boundedness of solutions of the problem (11). In addition of (9), assume that the Carath´eodory function a : Ω×IR×IR N → IRN satisfies for almost x in Ω, for every s ∈ IR and for every ξ, η ∈ IR N with ξ 6= η |a(x, s, ξ)| ≤ a0 (x) + |s|p−1 + |ξ|p−1 0

where a0 is a non negative function in Lp (Ω) with p0 =

(12) p p−1 ,

(a(x, s, ξ) − a(x, s, η)).(ξ − η) > 0.

and (13)

N

Let us denote by L p logα L the Orlicz space generated by the N -function N Θ(t) = t p logα (e + t). Assume that N

f ∈ L p logα L with α >

N (p − 1) . p

(14)

Our first result is the following Theorem 2.1 (Benkirane-Youssfi-Meskine,3 2008). Suppose that (9), (12), (13) and (14) are filled. Then, the problem (11) has at least one solution u in W01,p (Ω) ∩ L∞ (Ω) in the sense that Z Z a(x, u, ∇u) · ∇vdx = f vdx (15) Ω

Ω

for all v in D(Ω). Proof. We only give a sketch. Denote by Tk , the truncation at level k > 0, defined as Tk (s) = max(−k, min(s, k)) and set Gk (s) = s−Tk (s). To get rid of the lack of coerciveness of the operator A, we introduce the truncation operator. Let {fn } be a sequence of L∞ -functions such that fn → f in L1 (Ω) and |fn | ≤ |f |. The approximate problem  u ∈ W01,p (Ω) such that   Z  Zn a(x, Tn (un ), ∇un ) · ∇φ dx = fn φ dx  Ω   Ω 1,p holds for every φ in W0 (Ω)

(16)

(17)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

92

admits at least one solution thanks to the Leray-Lions existence theorem (see Leray-Lions14). For t > 0 and > 0, we use T (Gt (un )) as test function in (17). Being h decreasing, letting tends to 0+ we arrive at ! Z Z d p−1 p h (t) − |f |dx. (18) |∇un | dx ≤ dt {|un |>t} {|un |>t} On the other hand, H¨ older’s inequality enables us to get d − dt

Z

{|un |>t}

|∇un |dx ≤ (−µ0 (t))

d − dt

1 p0

Z

{|un |>t}

|∇un |p dx

! p1

,

(19)

where µn denotes the distribution function of un , that is µn (t) = |{x ∈ Ω : |un (x)| > t}|. Combining (18), (19) and the following well known inequality (see Talenti20 ) Z 1 1 d N CNN (µn (t))1− N ≤ − |∇un |dx, dt {|un |>t} we obtain h(t) ≤

p0

−µ0n (t)

Z

1

0

N p0 CNN (µn (t))p (1− N )

{|un |>t}

|f |dx

! pp0

.

In the above inequality, we use H¨ older’s inequality in Orlicz spaces, then we integrate both sides between 0 and τ obtaining by the definition of the rearrangement p0

p0

H(u∗n (σ)) ≤

2 p kf k p N L

p0

p

log α L

p0 N

N CN

Z

1 σ 1 |Ω|

ds 0

p s log α N

(e + s)

.

Thus, by (14), the sequence {un } is uniformly bounded in L∞ (Ω). It is now easy to get an estimation of {un } in the energy space W01,p (Ω) which implies that there is a function u such that un → u

weakly in W01,p (Ω) and a.e in Ω.

Then, we prove that ∇un → ∇u

a.e in Ω.

Which enables us to pass to the limit in (17) to get (15).

(20)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

93

3. Nonstandard growth Here, we do not assume the ∆2 -condition on the N-function M. In particular, the corresponding Orlicz-Sobolev spaces are not reflexive. We extend the class of functions satisfying (6) to the Orlicz-Sobolev setting by assuming one of the following two assumptions: Either f ∈ LN (Ω),

(21)

or m

f ∈ L (Ω) and

Z

+∞ ·

t M (t)

Nm −m

dt < +∞.

(22)

In addition of (9) and (13), we make the following assumption: |a(x, s, ξ)| ≤ a0 (x) + k1 P

−1

M (k2 |s|) + k3 M

−1

M (k4 |ξ|)

(23)

where a0 (x) belongs to EM (Ω), P is an N-function such that P M and ∗ k1 , k2 , k3 , k4 are constants in IR+ . We prove the following Theorem 3.1 (Youssfi,22 2007). Suppose that (9), (13) and (23) are filled. Under either (21) or (22), the problem (11) has at least one solution u in W01 LM (Ω) ∩ L∞ (Ω) in the sense that Z Z a(x, u, ∇u) · ∇vdx = f vdx (24) Ω

Ω

for all v in D(Ω). Remark 3.1. The previous theorem is surprising, since the function h does not influence the result. This seems to be natural, since if one looks for bounded solutions the degeneracy of the operator A ”disappears”. Our result includes classical similar results, notably the G. Stampacchia’s one. We extend the result of the previous theorem 3.1 to the strongly nonlinear problem ( A(u) + H(x, u, ∇u) = f in Ω, (25) u=0 on ∂Ω, where H is a Carath´eodory function which doesn’t satisfy necessarily the famous sign condition H(x, s, ξ)s ≥ 0, but only the following growth condition |H(x, s, ξ)| ≤ β(s)M (|ξ|),

(26)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

94

where β : IR → IR+ is a continuous function such that the function β(t)

t →

M

−1

(M (h(|t|)))

1

belongs to L (IR). So by defining Z s γ(s) = 0

β(t) M

−1

(M (h(|t|)))

dt

for all s ∈ IR, the function γ is bounded. Theorem 3.2 (Benkirane-Youssfi,4 2008). Suppose that (9), (13), (23) and (26) are filled. Under either (21) or (22), the problem (25) has at least one solution u in W01 LM (Ω) ∩ L∞ (Ω) in the sense that Z Z Z f vdx (27) H(x, u, ∇u)vdx = a(x, u, ∇u) · ∇vdx + for all v ∈

Ω

Ω

Ω

W01 LM (Ω)

∞

∩ L (Ω).

Remark 3.2. In addition of the degeneration of the operator A, the fact that H does not satisfy the sign condition creates another problem of getting the a priori estimates. To overcome this hindrance, we use test functions of exponential type containing the function γ. Note that the result of Theorem 3.2 extends those obtained in Boccardo et al.9 and Trombetti.21 Proof. Summarized. Step 1: A priori estimates: We define An and Bn as An (u) := −div a(x, Tn (u), ∇u) and Bn (x, u, ∇u) = Tn (B(x, u, ∇u)). Denote by m∗ either N or m according as we assume (21) or (22), and let {fn } ⊂ W −1 EM (Ω) be a sequence of smooth functions such that ∗

fn → f strongly in Lm (Ω)

and kfn km∗ ≤ kf km∗ . By Proposition 2 of Gossez-Mustonen,12 there exists at least one solution un ∈ D(An + Bn ) ⊂ W01 LM (Ω) to the approximate equation −div a(x, Tn (un ), ∇un ) + Bn (x, un , ∇un ) = fn

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

95

in the sense that Z Z Z fn vdx Bn (x, un , ∇n )vdx = a(x, Tn (un ), ∇un ) · ∇vdx +

(28)

Ω

Ω

Ω

for all v ∈ W01 LM (Ω).

∗ Lemma 3.1. Let un be a solution of (28). For all t, in IR+ , one has the following inequalities: Z + a(x, Tn (un ), ∇un ) · ∇un eγ(un ) dx {t
≤ Z

Z

(29)

+

{un >t}

fn+ eγ(un ) T (Gt (u+ n ))dx. −

{−t−
≤

Z

a(x, Tn (un ), ∇un ) · ∇un eγ(un ) dx

(30)

−

{un <−t}

fn− eγ(un ) T (Gt (u− n ))dx.

Proof. To prove (29) we use the test function +

eγ(Tk (un )) T (Gt (Tk (u+ n ))) that belongs to W01 LM (Ω) for all k > 0 and we use the function −

−eγ(Tk (un )) T (Gt (Tk (u− n )))

which belongs to W01 LM (Ω) to prove (30).

Lemma 3.2. There exists a constant c0 , not depending on n, such that for almost every t > 0 Z Z d −1 |fn |dx. (31) M (M (h(|un |)))M (|∇un |)dx ≤ c0 − dt {|un |>t} {|un |>t} +

−

Proof. The two functions eγ(un ) and eγ(un ) are bounded in L∞ (Ω), we sum up both inequalities (29) and (30) obtaining (31). The following comparison lemma constitutes the crucial step in the proof. Lemma 3.3. Let K(t) = Mt(t) and µn (t) = |{x ∈ Ω : |un (x)| > t}|, for all t > 0. We have for almost every t > 0: h(t) ≤ 2M (1)(−µ0n (t)) M

−1

1 N

(M (1))N CN µn (t)

1 1− N



 K −1  

c0 M

−1

Z

{|un |>t}

|fn |dx 1 N

(M (1))N CN µn (t)

1 1− N



 (32)  

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

96

where CN stands for the measure of the unit ball in IR N and c0 is the constant which appears in (31). 1 is decreasing and convex (see TalK −1 (t) 19 enti ). Hence, Jensen’s inequality yields  Z −1 M (M (h(|un |)))M (|∇un |)dx   {t<|un |≤t+k}  Z C   −1 M (M (h(|un |)))|∇un |dx Proof. The function C(t) =

{t<|un |≤t+k}

Z

 =C  ≤Z ≤

Z

{t<|u Zn |≤t+k}

M

{t<|un |≤t+k}

M {t<|un |≤t+k}

M {t<|un |≤t+k}

M

K(|∇un |)M

−1

−1

−1

−1

(M (h(|un |)))|∇un |dx

(M (h(|un |)))|∇un |dx

   

(M (h(|un |)))dx

(M (h(|un |)))|∇un |dx

M

−1

−1

(M (h(t + k)))

(M (h(t)))(−µ Z n (t + k) + µn (t)) {t<|un |≤t+k}

−1

.

|∇un |dx −1

Taking into account that M (M (h(t))) ≤ M (M (1)), using the convexity of C and then letting k → 0+ , we obtain for almost every t > 0   d Z −1 − M (M (h(|un |)))M (|∇un |)dx −1   dt {|un |>t} M (M (1))  Z C −1   d −1 M (M (h(t))) |∇un |dx) M (M (1))(− dt {|un |>t} 0 −µn (t) Z ≤ . d |∇un |dx − dt {|un |>t} Recall the following inequality, (see Talenti19 ): Z 1 d 1− 1 − |∇un |dx ≥ N CNN µn (t) N for almost every t > 0. dt {|un |>t}

(33)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

97

The monotonicity of the function C, (31) and (33) yield 1 M ≤

−1

M

(M (h(t)))

−1

−µ0n (t)

1 N

(M (1))N CN µn (t)

1 1− N

Using the inequality M (t) ≤ tM



 K −1  

−1

c0 M

−1

Z

{|un |>t}

|fn |dx 1 N

(M (1))N CN µn (t)

1 1− N



 . 

(M (t)) for all t ≥ 0

(34)

and the fact that 0 < h(t) ≤ 1, we obtain (32). Step 2: L∞ -bound: If we are under (21) we get kun k∞ ≤ H

2M (1)

−1

M

−1

1

K

−1

(M (1))N CNN

M

−1

c0 kf kN

1

(M (1))N CNN

!

N |Ω|

1 N

!

(35)

and if we are under (22) we get kun k∞ ≤ H

−1

(M

−1

2M (1)cr0 kf krm

r+1

(M (1)))r+1 N r CNN

K −1 (λ) + λr

Z

+∞ K −1 (λ)

s M (s)

r

ds

!!

.

(36)

This, implies that kun k∞ ≤ c.

(37)

Step 3: Estimation in W01 LM (Ω). An estimation in W01 LM (Ω), is obtained by proving the following Lemma 3.4. Z (1) −

{0≤un }

(2) −

Z

+

a(x, Tn (un ), ∇un ) · ∇un eγ(un ) dx ≤ −

{un ≤0}

a(x, Tn (un ), ∇un ) · ∇un eγ(un ) dx ≤

Proof. We use +

1 ∞ eγ(un ) u+ n ∈ W0 LM (Ω)∩L (Ω)

Z Z

+

Ω

fn+ eγ(un ) u+ n dx. −

Ω

fn− eγ(un ) u− n dx.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

98

to prove the first inequality and use −

1 ∞ −eγ(un ) u− n ∈ W0 LM (Ω)∩L (Ω)

for the second. Summing up both inequalities (1) and (2) we get Z 1 cc1 kf km∗ |Ω|1− m∗ . M (|∇un |)dx ≤ −1 Ω M M (h(c))

(38)

Hence, there is a function u ∈ W01 LM (Ω) such that un * u weakly in W01 LM (Ω) for σ(ΠLM , ΠEM ),

(39)

un → u in EM (Ω) strongly and a.e. in Ω.

(40)

and

Step 4: Almost everywhere convergence of the gradients. We prove the following Lemma 3.5. The sequence {a(x, Tn (un ), ∇un )} is uniformly bounded in (LM (Ω))N . Then, we prove that ∇un → ∇u a.e. in Ω.

(41)

a(x, Tn (un ), ∇un ) * a(x, u, ∇u) weakly in (LM (Ω))N for σ(ΠLM , ΠEM ).

(42)

and

Step 5: Modular convergence of the gradients. We prove that un → u in W01 LM (Ω) for the modular convergence. Step 6: Equi-integrability of the non-linearities. We prove that Bn (x, un , ∇un ) → B(x, u, ∇u) strongly in L1 (Ω).

(43)

Step 7: Passage to the limit. Testing by v ∈ W01 LM (Ω)∩L∞ (Ω), we have all ingredients to pass to the limit in (28) and obtain that u is a solution of (25) in the sense of (27).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

99

References 1. A. Alvino, L. Boccardo, V. Ferone, L. Orsina, G. Trombetti, Existence results for nonlinear elliptic equations with degenerate coercivity, Ann. Mat. Pura Appl., 182, (2003), 53–79. 2. A. Alvino, V. Ferone, G. Trombetti, A priori estimates for a class of non uniformly elliptic equations, Atti Semin. Mat. Fis. Univ. Modena 46-suppl., (1998), 381–391. 3. A. Benkirane, A. Youssfi, D. Meskine, Bounded solutions for nonlinear elliptic equations with degenerate coercivity and data in an L log L, Bull. Belg. Math. Soc. Simon Stevin 15 (2008), 369–375. 4. A. Benkirane, A. Youssfi, Existence of bounded solutions for a class of strongly nonlinear elliptic equations in Orlicz-Sobolev spaces, AJMAA, Vol. 5, No. 1, Art. 7, (2008), 1–26. 5. Benkirane, A., Youssfi, A.: Regularity for solutions of nonlinear elliptic equations with degenerate coercivity, Ricerche Mat. 56, no.2, (2007), 241– 275. 6. L. Boccardo, H. Brezis, Some remarks on a class of elliptic equations with degenerate coercivity, Boll. Unione Mat. Ital., 6, (2003), 521–530. 7. L. Boccardo, A. Dall’Aglio, L. Orsina, Existence and regularity results for some nonlinear equations with degenerate coercivity, Atti Sem. Mat. Fis. Univ. Modena, 46-suppl., (1998), 51–81. 8. L. Boccardo, F. Murat, P.-L. Puel, L∞ estimate for some nonlinear elliptic partial differential equations and application to an existence result, SIAM J. Math. Anal., (2) 23, (1992), 326–333. ´ n, C. Trombetti, Bounded and unbounded 9. L. Boccardo, S. Segura de Leo solutions for a class of quasi-linear elliptic problems with a quadratic gradient term, J. Math. Pures Appl., 80, (2001), 919–940. 10. V. Ferone, M. R. Posteraro, J. M. Rakotoson, L∞ -estimates for nonlinear elliptic problems with p-growth in the gradient, J. of Inequal. Appl., 3, (1999), 109–125. 11. E. DeGiorgi, Sulla differenziabilit` a e l’analiticit` a delle estremali degli integrali multipli regolari, Mem. Accad. Sci. Torino, Cl. Sci. Fis. Mat. Nat. 3 (1957), 25–43. 12. J.-P. Gossez and V. Mustonen, Variational inequalities in Orlicz-Sobolev spaces, Nonlinear Anal., 11, (1987), 379–49. 13. O. A. Ladyzhenskaya, N. Ural’tseva, Linear and Quasilinear Elliptic Equations, Academic Press, 1968. 14. J. Leray, J. L. Lions, Quelques r´esultats de Viˇsik sur les probl`emes elliptiques non lin´eaires par les m´ethodes de Minty-Browder, Bull. Soc. Mat. France 93, (1965), 97–107. 15. G. Stampacchia, Le probl`eme de Dirichlet pour les ´equations elliptiques du second ordre a ` coefficients discontinus, Ann. Ist. Fourier (Grenoble), 15, (1965), 189–258. 16. G. Stampacchia, ´equations elliptiques du second ordre a ` coefficients discontinus, Montr´eal, Presses Univ. Montr´eal, 1966, (S´eminaire de Math. sup., 16).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

100

17. G. Talenti, Inequalities in rearrangement invariant function spaces, Nonlinear analysis, function spaces and applications,Vol. 5. Proceedings of the spring School held in Prague, May 23-28, 1994, 177-230. Mathematical Institute, Czech Academy of Sciences, and Prometheus Publishing House, Praha 1995. 18. G. Talenti, Linear elliptic P.D.E’s: Level sets, rearrangements and a priori estimates of solutions, Boll. Un. Mat. Ital. 4-B(6), (1985), 917–949. 19. G. Talenti, Nonlinear elliptic equations, Rearrangements of functions and Orlicz spaces, Ann. Mat. Pura Appl., 120, 1979, IV. Ser., 159–184. 20. G. Talenti, Elliptic equations and rearrangements, Ann. Scuola. Norm. Sup. Pisa (4) 3, (1976), 697–718. 21. C. Trombetti, Non-uniformly elliptic equations with natural growth in the gradient, Potential Analysis, 18, (2003), 391–404. 22. A. Youssfi, Existence of bounded solutions for nonlinear degenerate elliptic equations in Orlicz spaces, Electron. J. Diff. Eqns., 2007, No. 54, (2007), 1–13.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

101

Analysis of a new mixed formulation of the obstacle problem Belkassem Seddoug, Ahmed Addou, and Abdeluaab Lidouh D´ epartement de Math´ ematiques et Informatique, Facult´ e des sciences Universit´ e Mohammed Premier, Oujda, Morocco For the obstacle problem:

∆u = f χ[u>ψ] + ∆ψ, u ≥ ψ,

for which the unknowns are u and the free boundary ∂[u > ψ]. It is known1 that the coincidence set [u = ψ] is contained in the set [f − ∆ψ ≥ 0]. In this work we introduce µ = (f − ∆ψ)+ χ[u>ψ] , that characterizes the domain of non contact,2 and we analyze a mixed formulation of the obstacle problem where µ appears as a Lagrange multiplier. We show, in particular, that the solution µ h of the approached (by linear finite element) mixed problem converges to µ with an order of convergence that is an O(h), which is optimal seen the equivalent result on the approximation of the free boundary 3 and.4 Keywords: Obstacle problem; Variational inequalities; Free boundary; Mixed formulation.

1. Introduction In this work we are interested in the following “model” obstacle problem: 1 Find u ∈ K = v ∈ H (Ω) : v ≥ ψ a.e. in Ω such that 0 R R (1) Ω ∇u (∇v − ∇u) dx + Ω f (v − u) dx ≥ 0, for every v ∈ K,

where ψ, f and Ω are regular data.

As in,1 problem (1) can be written: ∆u = f − µs in Ω, u|∂Ω = 0,

(2)

where µs is a positive measure whose support is contained in the coincidence set I(u) = {x ∈ Ω : u(x) = ψ(x)} .

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

102

The a priori determination of µs would permit to solve (1) as a Dirichlet problem. In the same sense, it has been established in,2 that problem (1) is equivalent to the following one:  1 2  Find u ∈ H0 (Ω) and µ ∈ L (Ω) such that − 1 (3) ∆u = µ + F + ∆ψ in H0 (Ω) ,  µ ∈ ∂ϕ(u), R where F = f − ∆ψ supposed in L2 (Ω), and ϕ : v 7−→ Ω F + v + dx. With h+ = max(h, 0), h− = min(h, 0) for every h in L2 (Ω), so that h = h+ + h− . ∂ϕ(u) designates the subdifferential of ϕ at u. A fundamental aspect, from that point of view, is the characterization of the non contact domain Ω+ (u) = Ω\I(u), and therefore the free boundary, which is also an unknown of the problem, with the help of µ. What we propose through this article that is the analysis of the approach presented in2 by putting it in a setting of mixed formulation where the measure µ appears as a Lagrange multiplier. It allows us to adapt the theory of Brezzi6 and7 for the analysis of the approximation by finite element of the reformulated problem. This point of view is different from the one presented in2 since one determines the two unknowns of the obstacle problem at the same time, u and µ (and so the free boundary), and also different from the one adopted by the different mixed formulations of the obstacle problem. 2. Survey of the abstract problem 2.1. The continuous problem Let us consider two Hilbert spaces H and V such that V ⊂ H ⊂ V 0 , with continuous and dense injections. We denote h., .i the duality pairing between V and V 0 , a(., .) the scalar product of V and (., .) the one of H. And we will note A the Riesz operator between V and V 0 (i.e. ∀(u, v) ∈ V 2 : a(u, v) = hAu, vi). We consider a closed convex cone K in H, its polar cone is defined by: K 0 = {w ∈ H such that ∀v ∈ K : (v, w) ≤ 0} .

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

103

If v ∈ H, v + and v − designate respectively the orthogonal projections of v on K and K 0 , i.e. v = v + + v − , (v + , v − ) = 0 and ∀w ∈ K : (v − v + , w − v + ) ≤ 0. We suppose in addition that: (H1) K is right i.e. K 0 = −K. (H2) ∀v ∈ V : v + , v − ∈ V and a(v + , v − ) = 0. (H3) ∀u, v ∈ V : u+ + v + − (u + v)+ ∈ K ∩ V. We consider the following variational inequality problem: Find u ∈ K ∩ V such that a(u, v − u) + (f, v − u) ≥ 0, for every v ∈ K ∩ V,

(4)

where f ∈ H. It is known that problem (4) admits a unique solution.1 2.2. Reformulation of the problem As in2 we consider, on V, the continuous and convex mapping ϕ : V −→ IR, v 7−→ ϕ(v) = (f + , v + ). The following propositions have been shown in.2 Proposition 2.1. Problem (4) is equivalent to the problem Find u ∈ V such that a(u, v − u) + ϕ(v) − ϕ(u) + (f − , v − u) ≥ 0, for every v ∈ V.

(5)

Problem (5) is different from (4) by the fact that it is without constraints. In addition, formulation (5) allows us to show the following continuity result. Lemma 2.1. if u is the solution to problem (5) then the linear form, w 7−→ a(u, w), is continuous on V for the norm of H. And one has the bound of the norm of u in H :

a(u, w) ≤ f + H + f − H . w∈V,w6=0 kwkH sup

(6)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

104

Proof. For every w in V , with v = u + w in (5) and thanks to (H3), one has: a(u, w) ≥ f + , (u+ + w+ ) − (u + w)+ − f + , w+ − (f − , w) ≥ − f + , w+ − (f − , w)

≥ − f + + f − kwk . H

H

H

And with v = u − w, one has:

a(u, −w) ≥ − f + H + f − H k−wkH ,

that is to say

a(u, w) ≤ f + H + f − H kwkH .

That finishes the proof.

Proposition 2.2. The function u is the solution to problem (5) if and only if (u, µ) is the solution to problem   Find (u, µ) ∈ V × H such that a(u, v) + (µ, v) + (f − , v) = 0, for every v ∈ V, (7)  µ ∈ ∂ϕ(u),

∂ϕ(u) = {λ ∈ H | ∀w ∈ H : ϕ(u) − ϕ(w) ≤ (λ, u − w)} designates the subdifferential of ϕ at u.

Taking into account the following lemma 2.2 (Lemma 1 in2 ), problem (7) can be written as:   Find (u, µ) ∈ V × C such that (8) a(u, v) + hµ, vi + (f − , v) = 0, for every v ∈ V,  hζ − µ, ui ≤ 0, for every ζ ∈ C,

where C = {ζ ∈ H : hζ, vi ≤ ϕ(v), for every v ∈ V } is a closed convex containing 0H . Lemma 2.2. For every µ ∈ H the following propositions are equivalent (i) µ ∈ ∂ϕ(u). (ii) µ ∈ C and hµ, ui = ϕ(u). (iii) µ ∈ C and hζ − µ, ui ≤ 0 for every ζ ∈ C.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

105

In,2 for the calculation of µ then the resolution of a Dirichlet problem to determine u, the authors proposed to uncouple problem (8) by putting z = A−1 (µ) and t = A−1 (f − ). Problem (8) is then written:   Find (u, z) ∈ V × M such that a(u + z + t, v) = 0, for every v ∈ V,  a(w − z, u) ≤ 0, for every w ∈ M,

where A is the Riesz-Fr´echet operator associated to a(., .) and M = A−1 (C). That permits to define z as being the projection of −t on the closed convex M for the scalar product a(., .). And in order to compute z, they proposed a projection algorithm inspired from the one of Degueil.8 The set M being defined only implicitly, so the proposed algorithm cannot be implemented, and makes the analysis of the discrete problem nearly impossible. In addition to calculate t it is necessary to solve the problem At = f − in advance. What we propose, in alternative, it is the direct resolution (through an approximation by finite element) of the mixed problem (8), especially as the convex C can be determined explicitly, more precisely one has. Lemma 2.3. With the hypotheses above one has C = f + + K 0 ∩ K.

Proof. By definition of C and ϕ, one has, for every ζ ∈ C ∀v ∈ V : (ζ, v) ≤ (f + , v + ).

The space V being dense in H and the projection operator from H on the cone K is continuous, so the previous inequality is verified for all v in H. While taking v respectively in K and K 0 , one gets ∀v ∈ K : ζ − f + , v ≤ 0 and ∀v ∈ K 0 : (ζ, v) ≤ (f + , v + ) = 0, 0 then ζ − f + ∈ K 0 and ζ ∈ K 0 = K. On the other hand if ζ ∈ K and ζ − f + ∈ K 0 , one has, for all v ∈ V (ζ, v) = ζ, v + ) + (ζ, v − ≤ ζ, v + ≤ (f + , v + ), so ζ ∈ C.

Remark 2.1. While the convex C is not a cone, one cannot apply directly the theory of Brezzi.6 In this case an alternative is proposed in7 to justify

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

106

existence and uniqueness of the solution to problem (8) through an inf-sup condition between the spaces V and H equipped with the norm of V 0 , which is obvious in our case, indeed, for every α ∈ V 0 one has: hα, vi = kαkV 0 . v∈V,v6=0 kvkV sup

(9)

Remark 2.2. The bilinear form a being symmetric, so the mixed problem (8) is equivalent to the following saddle point problem:9 Find (u, µ) ∈ V × C such that (10) L(u, ζ) ≤ L(u, µ) ≤ L(v, µ) , for all (v, ζ) ∈ V × C, where the Lagrangian L is defined on V × V 0 by L(v, ζ) = (f − , v) + hζ, vi.

1 a(v, v) + 2

2.3. Property of stability Seen should be given the inf-sup condition (9) problem (8) can be studied regardless of problem (4) (see7 ), so we have the following stability result. Proposition 2.3. There exists a positive constant C such that

kukV + kµkV 0 ≤ C f − H ,

(11)

where (u, µ) denote the solution of problem (8), and k.kX the norm of the Hilbert space X. Proof. With v = −u and ζ = 0 in (8), one has

a(u, −u) + h−µ, ui + (f − , u) = 0 and h−µ, ui ≤ 0,

hence

kukV ≤ c f − H .

(12)

To estimate kµkV 0 , we also use (8), so for all v in V one has: hµ, vi = −(f − , v) − a(u, v)

≤ f − H kvkH + kukV kvkV ≤ C f − H + kukV kvkV ,

that implies with (12) the estimation (11).

Remark 2.3. The stability result above shows how much f − controls the solution of the problem. in particular if f − = 0 then u = µ = 0.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

107

Remark 2.4. If the constraint of problem (4) is given by the translated w + K of K by an element w ∈ W (with V ⊂ W ⊂ H such that the bilinear → − → − form a is defined and continuous on W ), problem (5) can be written Find u ∈ V such that (13) a(u − w, v − u) + ϕ1 (v) − ϕ1 (u) + (F − , v − u) ≥ 0, for all v ∈ V, where ϕ1 : v 7−→ (F + , (v − w)+ ) and F defined by (F, v) = (f, v) + a(w, v) for all v ∈ V, we suppose that F is continuous for the norm of H. Problem (8) becomes:  0  Find (u, µ) ∈ V × C such that (14) a(u − w, v) + hµ, vi + (F − , v) = 0, for every v ∈ V,  hζ − µ, u − wi ≤ 0, for every ζ ∈ C 0 , where C 0 = F + + K 0 ∩ K. 2.4. The discrete problem Let (Vh )h be a sequence of finite dimensional subspaces of V , and for all h we define Vh0 as the dual of Vh by Vh0 = {a(vh , .) such that vh ∈ Vh } = {(vh , .) such that vh ∈ Vh } ⊂ V 0 , subspace of V 0 and of H (Vh being of finite dimension so the restrictions to Vh of the two norms k.kV and k.kH are equivalent). We consider, finally, a sequence (Ch ) of closed convex subsets of Vh containing 0V 0 . And we assume that the sequence (Vh , Ch ) approximates (V, C) in the sense of.5

(15)

The discrete problem associated to (8) can be written as:   Find (uh , µh ) ∈ Vh × Ch such that a(uh , vh ) + hµh , vh i + (f − , vh ) = 0, for all vh ∈ Vh ,  hζh − µh , uh i ≤ 0, for all ζh ∈ Ch .

(16)

As in,7 we introduce the following discrete inf-sup condition: ∀αh ∈ Vh0 :

hαh , vh i ≥ β kαh kV 0 , vh ∈Vh ,vh 6=0 kvh kV sup

(17)

for some β > 0 independent of h. What permits to show (see7 ) the following stability result.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

108

Proposition 2.4. If the assumption (17) holds then there exists a constant c such that

kuh kV + kµh kV 0 ≤ c f − H , where (uh , µh ) denotes the solution of problem (16).

In order to show a convergence result, first one has to show the following estimate, necessary especially for the error estimate. Proofs can be found in7 and.10 Theorem 2.1. Under assumption (17), if (u, µ) and (uh , µh ) denote the respective solutions to problems (8) and (16), then there exists a constant c independent of h such that the following estimates hold: ku − uh k2V ≤ c inf ku − vh k2V + inf kµ − ζh k2V 0 + B1 (ζh ) ζh ∈Ch vh ∈Vh (18) + inf B2 (ζ) , ζ∈C

kµ −

µh k2V 0

≤ c ku −

uh k2V

+ inf kµ − ζh ∈Ch

ζh k2V 0

,

(19)

where B1 (ζh ) = hµ − ζh , ui and B2 (ζ) = hµh − ζ, ui. We finally state the convergence result. Theorem 2.2. Under assumptions (15) and (17), if (u, µ) and (uh , µh ) denote the respective solutions to problems (8) and (16), then (u h , µh ) converges strongly to (u, µ) in V × V 0 . 3. Study of an obstacle problem 3.1. The continuous problem Let Ω be a bounded domain (that one supposes polygonal to simplify the analysis) in Rn (n ≤ 2). The Hilbert space H01 (Ω) is equipped with the scalar product: Z a(u, v) = ∇u∇vdx, for all (u, v) ∈ H01 (Ω). Ω

Let f ∈ H that

−1

(Ω) and ψ ∈ H 1 (Ω) such that ψ ≤ 0 on ∂Ω, we suppose F = (f − ∆ψ) ∈ L2 (Ω),

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

109

and we define Kψ = v ∈ L2 (Ω) : v ≥ ψ a.e. in Ω .

We consider the following obstacle problem: Find u ∈ Kψ ∩ H01 (Ω) such that a(u, v − u) + (f, v − u) ≥ 0, for all v ∈ Kψ ∩ H01 (Ω).

(20)

The hypotheses of the section (2) are verified for V = H01 (Ω), W = H (Ω), H = L2 (Ω), and Kψ = ψ + K (w = ψ), with K the convex cone of L2 (Ω) of almost everywhere positive functions in Ω i.e. K = −K 0 = v ∈ L2 (Ω) : v ≥ 0 a.e. in Ω . 1

Problem (20) is then equivalent to the following one:  1 0  Find (u, µ) ∈ H0 (Ω) × C such that (21) a(u − ψ, v) + (µ, v) + (F − , v) = 0, for all v ∈ H01 (Ω),  (ζ − µ, u − ψ) ≤ 0, for all ζ ∈ C 0 , with C 0 = K ∩ F + + K 0 = ζ ∈ L2 (Ω) such that 0 ≤ ζ ≤ F + a.e. in Ω . On the other hand if (u, µ) is solution to problem (21) then u is the solution to the following Dirichlet problem: ∆u = µ + F − + ∆ψ in Ω, (22) u|∂Ω = 0. In addition it is known1 that the coincidence set [u = ψ] is contained in the set [F ≥ 0] and ∆(u − ψ) = F χ[u>ψ] ,

(23)

F − χ[u=ψ] = 0,

(24)

what means:

and ∆u = F + χ[u>ψ] +F − {1−χ[u=ψ] }+∆ψ = F + χ[u>ψ] +F − +∆ψ in Ω. (25) Combining (22) and (25) we can write µ = F + χ[u>ψ] a.e in Ω.

(26)

Remark 3.1. In order to simplify the analysis, we will suppose that ψ = 0. Seen should be given the form of problem (21), it is clear that this hypothesis is not restrictive.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

110

3.2. The discrete problem Let us consider that a triangulation Th is defined over Ω, regular in the sense that each simplex T ∈ Th contains a ball with radius γ1 h and is contained in a ball with radius γ2 h where the positive constants γ1 and γ2 are independent of h. A piecewise linear subspace Vh of V can be defined as: ¯ ∩ H 1 (Ω) such that for all T ∈ Th : vh|T ∈ P1 (T ) . Vh = v h ∈ C 0 Ω 0

Let N = N (h), number of nodes Ai of the triangulation, to simplify the notations, we assume that the numbering is such that A1 , A2 , . . . , AN0 are the internal nodes while AN0 +1 , . . . , AN lie on the boundary Γ. N0 is then the dimension of Vh . We introduce the canonical basis (ϕ1 , ϕ2 , . . . , ϕN ) associated to the triangulation Th . We assume in addition that the inverse assumption (see11 ) holds, what permits to have the following inverse inequality: kwh kH 1 (Ω) ≤ ch−1 kwh kL2 (Ω) , ∀wh ∈ Vh .

(27)

0

To approximate µ ∈ L2 (Ω), we need a discrete space with the same dimension as Vh , so we can use Vh or we associate to the triangulation Th the set Kh of the N volumes Di (i = 1...N ) that constitute the dual of the triangulation Th known as the Voronoi mesh. This mesh is constructed by connecting with a straight line segment the mid-points of edges and centroids of each neighboring pair of triangles having a common edge. And we introduce the space Wh (approximation of L2 (Ω) and H −1 (Ω)), defined by: Wh = wh ∈ L2 (Ω) such that for all D ∈ Kh : wh|D ∈ P0 (D) ,

equipped with its canonical basis (χ1 , χ2 , ..., χN0 ), where χi is the characteristic function of the control volume Di . The discrete problem associated to (21) reads:   Find (uh , µh ) ∈ Vh × Ch such that a(uh , vh ) + hµh , vh i + (f − , vh ) = 0, for all vh ∈ Vh ,  hζh − µh , uh i ≤ 0, for all ζh ∈ Ch ,

where we set ( Ch =

ζh =

and Ti+ =

R

Ω

N0 X i=1

ζi χi ∈ Wh such that 0 ≤ hζh , ϕi i ≤

f + ϕi dx.

Ti+

: i = 1 . . . N0

(28)

)

,

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

111

With the notations above it is easy to see that problem (28 ) can be written as: ( QN 0 0, Ti+ such that Find (U, Y ) ∈ RN0 × i=1 Q N0 0, Ti+ , J (U, z) ≤ J (U, Y ) ≤ J (w, Y ), for all (w, z) ∈ RN0 × i=1 (29) 1 − (M w, w) + (T − , w) + (z, w) with Ti = where we set J (w, z) = 2 R − Ω f ϕi dx and Y = P µh . M is the stiffness matrix of the problem i.e. M = Z(mij = a(ϕZi , ϕj )), P = (pij ), is the N0 × N0 -matrix such that pij =

ϕi χj dx =

Ω

ϕi dx. In the one dimensional case and with regu-

Dj

lar triangulation, the matrix P is easily calculable and one has: 

6 1  .. 1 . h  P =  0 ... 8 . .  .. . . 0 ···

0 ··· .. .. . . .. .. . . .. .. . . 0 1

 0 ..  .   . 0   1 6

In the two dimensional case the matrix P is also calculable for a triangulation with simple geometries and it is symmetric. More precisely one 11 can show, in an elementary manner, the following properties: pii = |Di | 18 Z X 7 |Di | for i = 1...N0 . To compute pij = and pij = φi dx for i = j 18 Dj j6=i

or for internal neighboring nodes Ai and Aj , one can use trapezoid formula by splitting Dj into small triangles and using the fact that the dual mesh splits any triangle of the primal triangulation into six small triangles having common measure. If one uses Vh instead of Wh Zto approach µ, the matrix P will be replaced by B = (bij ), where bij =

ϕi ϕj dx. Using Simpson’s formula, one X 1 1 bij = |Ki |, where can show the following properties: bii = |Ki | and 3 6 j6=i [ Ki = T , for all internal node Ai , i = 1...N0 . Ω

Ai ∈T

To use the results of the section (2) we must establish a discrete inf-sup condition and show that Ch approximates C in the sense of.5 To this end, we establish the following technical lemmas.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

112

Lemma 3.1. There exists two positive constants c1 , c2 independent of h such that c1 kwh k2L2 (Ω) ≤ kαh k2L2 (Ω) ≤ c2 hαh , wh i , P P for all αh = αi χi ∈ Wh , where wh = αi ϕi ∈ Vh . i

(30)

i

Proof. In 2-D case. P P For α = t (α1 , ..., αN ), αh = αi χi ∈ Wh and wh = αi ϕi ∈ Vh , one i

has

2

kαh kL2 (Ω) = t αDα,

i

2

kwh kL2 (Ω) = t αBα, and hαh , wh i = t αP α,

where D is the diagonal matrix D = diag(|D1 | , ..., |DN |). So kwh k2L2 (Ω) kαh k2L2 (Ω)

t

=

t αBα t αBα t αα = ≤ , t αDα t αα t αDα s

(31)

where t is the largest eigenvalue of B, and s the smallest one of D. By the regularity assumptions on the triangulation and the properties of B and D, and thanks to Gershgorin’s theorem, one has t≤

1 max |Ki | ≤ ch2 and s ≥ min |Di | ≥ ch2 , i 2 i

the first inequality of (30) follows. On the other hand 2

kαh kL2 (Ω) hαh , wh i

t

=

t αDα αDα t αα t0 = ≤ , t αP α t αα t αP α s0

(32)

where t0 is the largest eigenvalue of D, and s0 the smallest one of P. While reasoning like previously, the second inequality of (30) follows. Lemma 3.2. For λ ∈ L2 (Ω), if λI is the projection of λ on Wh in the following sense Z Z λvh dx = λI vh dx, ∀vh ∈ Vh , Ω

Ω

then there exists a positive constant c independent of h and λ such that

λ − λI −1 ≤ ch kλkL2 (Ω) . (33) H (Ω)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

113

Proof. For all v ∈ H01 (Ω), if v I ∈ Vh denotes the Lagrange interpolant of v, one has Z Z (λ − λI )vdx = (λ − λI )(v − v I )dx Ω

Ω

≤ λ − λI L2 (Ω) v − v I L2 (Ω)

≤ ch λ − λI L2 (Ω) kvkH 1 (Ω) , 0

therefore

λ − λ I

≤ ch λ − λI L2 (Ω) . (34) P P On the other hand, let λI = λi χi , and wh = λi ϕi , using (30 ), one has

I 2

λ 2

L (Ω)

≤c

Z

H −1 (Ω)

i

I

i

λ wh dx = c

Z

Ω

Ω

λwh dx ≤ c kλkL2 (Ω) λI L2 (Ω) ,

what implies, with the help of (34), the estimate (33).

Lemma 3.3. There exists a constant β independent of h, such that ∀αh ∈ Wh : Proof. For αh =

N P

i=1

such that

hαh , vh i ≥ β kαh kH −1 (Ω) . kv h kH 1 (Ω) vh ∈Vh ,vh 6=0 sup

(35)

0

αi χi ∈ Wh , let w ∈ H01 (Ω) ∩ H 2 (Ω) and wh ∈ Vh ,

−∆w = αh and a(wh , vh ) = a(w, vh ) = hαh , vh i , ∀vh ∈ Vh , it is known (see12 ), that kw − wh kH 1 (Ω) ≤ ch kαh kL2 (Ω) , therefore 0

2 kwkH 1 (Ω) 0

−

2 kwh kH 1 (Ω) 0

2

2

= kw − wh kH 1 (Ω) ≤ ch2 kαh kL2 (Ω) , 0

hence kαh k2H −1 (Ω) = kwk2H 1 (Ω) ≤ kwh k2H 1 (Ω) + ch2 kαh k2L2 (Ω) . 0

On the other hand, for vh =

0

N P

i=1

inequality (27), one has

αi ϕi ∈ Vh , using (30) and the inverse

hαh , vh i ≥ ch kαh kL2 (Ω) , kvh kH 1 (Ω) 0

(36)

(37)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

114

what permits to conclude, with the help of (36) that kαh kH −1 (Ω) ≤ c kwh kH 1 (Ω) = c 0

hαh , vh i . kv h kH 1 (Ω) vh ∈Vh ,vh 6=0 sup

0

If one uses Vh instead of Wh to approximate µ, the discrete inf-sup condition would be an immediate consequence of (36) and (27). Lemma 3.4. The sequence (Ch ) approximates C in the following sense: (i) ∀λ ∈ C, ∃λh ∈ Ch ,such that: λh → λ in H −1 (Ω). (ii) If λh ∈ Ch such that λh * λ in H −1 (Ω), for some λ in L2 (Ω), then λ ∈ C. Proof. Item (i) follows from the estimate (33) and the fact that λI ∈ Ch if λ ∈ C. On the other hand let λh and λ as in (ii), so for all v ∈ H01 (Ω), v ≥ 0 and vh ∈ Vh , vh ≥ 0 such that vh → v in H01 (Ω), one has Z Z f + vh dx, λh vh dx ≤ 0≤ however

R

Ω

λh vh dx →

R

Ω

Ω

λvdx, therefore Z Z 0≤ λvdx ≤ f + vdx,

Ω

Ω

Ω

what permits to conclude that λ ∈ C.

In addition to the results of Section 2, one has the following error estimate. Theorem 3.1. If (u, µ) and (uh , µh ) designate the respective solutions to problems (21) and (28), and if u ∈ H 2 (Ω) then there exists a positive constant c independent of h such that: ku − uh kH 1 (Ω) + kµ − µh kH −1 (Ω) ≤ ch. 0

(38)

Proof. Using theorem 2.1, and thanks to the interpolation error (see12 ):

u − u I 2 ≤ ch2 kukH 2 (Ω) and u − uI H 1 (Ω) ≤ ch kukH 2 (Ω) , L (Ω) 0

I

H01 (Ω),

where u ∈ Vh denotes the interpolant of u ∈ and the interpolation (33):

µ − µI −1 ≤ ch kµkL2 (Ω) , H (Ω)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

115

we had proved, it rests to examine the quantities B1 (ζh ) and B2 (ζ) that appear in the inequalities (18) and (19). To this end, let us take ζh = µI ∈ Ch in B1 , we have:

B1 (µI ) = µ − µI , u = µ − µI , u − uI

≤ µ − µI −1 u − uI 1 H

H0 (Ω)

(Ω)

2

≤ ch kµkL2 (Ω) kukH 2 (Ω) .

On the other hand, with ζ = µ ∈ C in B2 , and while taking into account the fact that u ≥ 0 a.e. in Ω and

hµ, ui = f + , u , we obtain:

B2 (µ) = hµh − µ, ui = hµh − f + , u − uI i + hµh − f + , uI i

≤ hµh − f + , u − uI i ( because µh ∈ Ch and uI ≥ 0)

≤ µh − f + L2 (Ω) u − uI L2 (Ω)

≤ µh − f + L2 (Ω) ch2 kukH 2 (Ω) ,

to conclude, we must show that kµh kL2 (Ω) is bounded independently of h, more precisely we show that

(39) kµh kL2 (Ω) ≤ c f + L2 (Ω) , P N0 P N0 indeed, let µh = i=1 µi χi and vh = i=1 µi ϕi , one has (lemma 3.1): Z N 0 X 2 kµh kL2 (Ω) ≤ c2 hµh , vh i = c2 µi µh ϕi dx i=1

≤ c2 ≤ c2

N0 X i=1

Z

|µi |

f Ω

+

Z

Ω

N0 X i=1

Ω

µh ϕi dx ≤ c2

N0 X i=1

|µi |

|µi | ϕi dx ≤ c2 f +

N 0

X

+

|µi | χi ≤ c f L2 (Ω)

2

i=1 L (Ω)

+ ≤ c f L2 (Ω) kµh kL2 (Ω) .

Z

f + ϕi dx Ω

N 0

X

|µ | ϕ

i i 2 L (Ω)

i=1

L2 (Ω)

That finishes the proof. In the case where one chooses Vh instead of Wh to approximate µ , the inequality (39) can be proved in the same manner by putting µh = PN0 i=1 µi ϕi , and using two times the lemma 3.1.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

116

3.3. Some numerical results In this section, we present two numerical tests in order to confirm the theoretical results. Example 1. We consider the following one-dimensional obstacle problem, with Ω = ]0, 2[, ψ = 0 and f defined by: f (x) = −1 if x ∈ [0, 1] , f (x) = 1 if x ∈ ]1, 2] . The solution to the continuous problem is calculable. And one has:  √ 1   u(x) = − x2 + (2 − 2)x if x ∈ [0, 1] ,   2 √ 1 2 √ u(x) = x − 2x + 1 if x ∈ 1, 2 ,     u(x) = 02 if x ∈ √2, 2 ,

and the measure µ = χ[1,√2] .

We solve the mixed problem (29) using Uzawa’s algorithm, and for different values of h. Figure 1 bellow represents the L2 -error on µ for different values of h.

0.15 0.14 0.13 0.12 0.11

L2-error on mu

0.1 0.09 0.08 0.07 0.06 0.05 0.04 0.03 0.02 0.01 0

0

0.01

0.02 h

Fig. 1.

L2 -error on µ for different values of h.

0.03

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

117

Figure 2 gives a comparison between the curve of u and the one of uh for h = 1/32. One remarks that the two curves are nearly confounded, and this even though h is not small enough.

0.8 0.7

Exact u. Approched u. h = 1 / 32.

0.6

u(x)

0.5 0.4 0.3 0.2 0.1 0

0

0.5

1

1.5

x

Fig. 2.

Comparison between u and uh for h = 1/32.

Figure 3 gives a comparison between the curve of µ and the one of µh for h = 1/256. The oscillations near the value 1 are due to the discontinuity of f at this point. The advantage to use µ to characterize the free boundary and not u, is due to the fact that with µ = F + χ[u>ψ] the jump is pronounced (because under some regularity hypotheses on the obstacle ψ (see13 ) one has F ≥ c > 0 near the free boundary), what is not the case for u that leaves the obstacle with a speed at least quadratic. It is what one observes well on the two figures 2 and 3.

Example 2. 2

We consider in the 2-D case, the following problem, with Ω = ]−1, 1[ , ψ = 0 and f as: f (x, y) = 8x2 + 8y 2 − 1.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

118 Approched mu. With h = 1 / 256. Exact mu. 1.1 1 0.9 0.8

mu(x)

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0

0.5

Fig. 3.

1 x

1.5

Comparison between µ and µh for h = 1/256.

The solution to the obstacle problem is the function u defined on Ω by: ( 1 u(x, y) = (4x2 + 4y 2 − 1)2 if 4x2 + 4y 2 − 1 ≤ 0, 32 u(x, y) = 0 if 4x2 + 4y 2 − 1 ≥ 0. We solve the mixed problem for different values of h, results obtained seem to be in agreement with the established theoretical results. Figures 4 and 5 represent respectively the isovalues of the functions u and uh for h = 1/30. We can easily notice the strong similarities between the two figures. In the same manner Figures 6 and 7 represent respectively the isovalues of exact µ and µh , the approximation of µ. for h = 1/60, we notice also the similarities of the two figures. Figure 8 represents the H01 (Ω)-error on u for different values of h, one can notice the linear aspect of this evolution, what confirms the error estimate of the theorem 3.1. Finally, we note that 2-D simulations have been realized under the environment FreeFem++ .14 Current and future developments The aim of this work was to present a new mixed formulation of the obstacle problem. We have proved that this formulation is as effective as those

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

119 u exact 60x60 IsoValue 0.00078125 0.00234375 0.00390625 0.00546875 0.00703125 0.00859375 0.0101563 0.0117188 0.0132813 0.0148437 0.0164063 0.0179687 0.0195313 0.0210938 0.0226562 0.0242188 0.0257812 0.0273438 0.0289063 0.0304687

Fig. 4.

Isovalues of exact u for h = 1/30.

classically known. Finally we have confirmed the theoretical results that we have established by numerical tests. Finally we would like to confirm that the approach we have presented can be used successfully for other similar problems, in particular the bilateral obstacle problem. Acknowledgement The authors would like to thank the anonymous referees for their interesting remarks. References 1. Kinderleher, D. Stampacchia. An introduction to variational inequalities and their applications. Academic Press (1980). 2. A. Addou, E. B. Mermri. Sur une m´ethode de r´esolution d’un probl`eme d’obstacle. Math-Recherche & Application n◦ 2, 59–69 (2000). 3. F. Brezzi & L. A. Caffarelli. Convergence of the discrete free boundaries for

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

120 uh 60x60 IsoValue 0.000676368 0.00223902 0.00380167 0.00536432 0.00692697 0.00848962 0.0100523 0.0116149 0.0131776 0.0147402 0.0163029 0.0178655 0.0194282 0.0209908 0.0225535 0.0241161 0.0256788 0.0272414 0.0288041 0.0303667

Fig. 5.

4. 5. 6. 7.

8.

9. 10.

Isovalues of uh for h = 1/30.

finite element approximations. R.A.I.R.O Numerical Analysis vol. 17, n◦ 4, 385–395 (1983). Ricardo H. Nochetto. Sharp L∞ -Error Estimates for semilinear Elliptic Problems with free boundaries. Numer. Maths. 54, 243-255 (1988). R. Glowinski, J. L. Lions, R. Tr´emoli`eres. Analyse num´erique des in´equations variationnelles Tome1. Dunod, Paris (1976). F. Brezzi, W. W. Hagar. Error estimate for the finite element solution of variational inequality, part II. Numer. Math. 31, 1–16 (1978). Slimane, L; Bendali, A; Laborde, P. Mixed formulations for a class of variational inequalities. M2AN Math. Model. Numer. Anal. 38 (2004), no. 1, 177–201. Degueil A. R´esolution par une m´ethode d’´el´ements finis d’un probl`eme de STEPHAN en terme de temp´erature et en teneur en mat´eriau non gel´e. Th`ese 3`eme cycle, Bordeaux (1977). I. Ekland, R. Temam. Analyse convexe et probl`emes variationnels. Dunod, Paris (1974). Slimane, L. M´ethodes mixtes et traitement du verrouillage num´erique pour la r´esolution des in´equations variationnelles. Th`ese de l’Institut National des Sciences Appliqu´ees, TOULOUSE (2001).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

121 mu exact 120x120 IsoValue 0.025 0.075 0.125 0.175 0.225 0.275 0.325 0.375 0.425 0.475 0.525 0.575 0.625 0.675 0.725 0.775 0.825 0.875 0.925 0.975

Fig. 6.

Isovalues of exact µ for h = 1/60.

11. P. G. Ciarlet. The Finite Element Method for Elliptic problems. Amsterdam: North Holland, (1978). 12. P. A. Raviart, J. M. Thomas. Introduction l’analyse num´erique des ´equations aux d´eriv´ees partielles. Masson, Paris (1992). 13. L. A. Caffarelli. The Obstacle Problem revisted. J. Fourier Anal. Appl. 4, 383–402 (1998). 14. F. Heicht, O. Pironneau, A. Le Hyaric, K. Ohtsuka. Freefem++. Laboratoire Jacques-Louis Lions, Universit´e Pierre et Marie Curie, Paris, (http://www.freefem.org/ff++).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

122

muh 120x120 IsoValue 0.0051434 0.0618342 0.118525 0.175216 0.231906 0.288597 0.345288 0.401979 0.458669 0.51536 0.572051 0.628742 0.685433 0.742123 0.798814 0.855505 0.912196 0.968886 1.02558 1.08227

Fig. 7.

Isovalues of µh for h = 1/60.

0.015

H1-error on u

0.01

0.005

0

0.05

0.1

0.15

0.2

h

Fig. 8.

The H01 (Ω)-error on u for different values of h.

recent

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

123

An obstacle problem via a sequence of penalized problems E. Azroul and M. Rhoudaf D´ epartement de Math´ ematiques et Informatique Facult´ e des Sciences Dhar-Mahraz B.P 1796 Atlas F` es, Morocco In this paper, we shall concern with the existence result of unilateral parabolic degenerated problems associated to the equations of the form ∂u + A(u) = f in QT , ∂t where A is a classical Leray-Lions operator acting from the weighted Sobolev 0 0 space Lp (0, T, W01,p (Ω, w)) into its dual Lp (0, T, W −1,p (Ω, w ∗ )), while the 1 datum f is assumed in L (QT ). The proof is based on the penalty methods. Keywords: Unilateral parabolic degenerate problem; Existence result; Penalty methods.

1. Introduction In this paper, we investigate the problem of existence of solutions of the obstacle problems associated to the following initial-boundary value problem:  ∂u  ∂t − div(a(x, t, u, ∇u)) = f in QT = Ω × (0, T ) (Pe ) u = 0 on Σ = ∂Ω × (0, T )  u(0) = u0 in Ω,

where Ω is an open bounded subset of IRN , N ≥ 1, T > 0, and we have set QT the cylinder Ω × (0, T ) and Σ its lateral surface. We assume that a(x, t, ξ) : QT × IRN → IRN is a Carath´eodry function (i.e., measurable with respect to (x, t) and continuous with respect to ξ) satisfying the hypotheses (H2 ) below. The data are taken such that: f ∈ L1 (QT ), u0 ∈ L1 (Ω) and u0 ≥ 0. More precisely, this paper deals with the existence of solution to the obstacle

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

124

degenerated parabolic problem (Pe ) in the sense of entropy solution:  u ≥ ψ a.e. in QT     Tk (u) ∈ Lp (0, T, W01,p (Ω, w)), u ∈ C([0, T ], L1 (Ω))  Z Z    ∂ϕ   S (u − ϕ)(τ ) dx + Tk (u − ϕ) dx dt k  Qτ ∂t Ω Z Z (Pu )   ≤ f T (u − ϕ) dx dt + Sk (u0 − ϕ(x, 0)) dx  k   Qτ Ω     ϕ ∈ Kψ ∩ L∞ (QT ) ∩ C([0, T ], L1 (Ω)) such that   ∂ϕ p0 −1,p0 (Ω, w∗ )), ∀ k > 0 ∂t ∈ L (0, T, W

where Sk (t) =

Z

t

Tk (s) ds,

0

ψ ∈ L∞ (Ω) ∩ W01,p (Ω, w) and Kψ = {u ∈

Lp (0, T, W01,p (Ω, w)), u ≥ ψ a.e. in QT }. The aim of our work is to investigate the relationship between the possibility to find solutions of (P ) by approximating the singular data f and u0 with sequences of regular functions. More precisely letting {f } and u0 be a standard approximation of f and u0 (that is f → f in L1 (Q) and u0 → u0 in L1 (Ω)), and considering the approximate problem:  ∂u 1 −  ∂t − div(a(x, t, u , ∇u )) − T (u − ψ) = f in QT (Pe ) u = 0 on Σ  u (0) = u0 in Ω.

We study the possibility to find a solution of (Pu ) as a limit of a subsequence {u} of solutions of (Pe ). The penalized term 1 T (u − ψ)− introduced in (Pe ) play a crucial role in the proof of our main result, in particular this term allows to prove that the solution u of (Pu ) belongs in Kψ (that is u ≥ ψ). A priori estimates of the truncations Tk (u ) are obtained in Lp (0, T, W01,p (Ω, w)). For the passage to the limit, we prove the strong converge of the truncation of u and the almost everywhere convergence of ∇u is proved. An example of operator model is, A(u) = −div(|x|r |∇u|p−2 ∇u), r > 0. In this context of parabolic problems, if w ≡ 1, existence results for (Pe ) 0 have been proved in18 when f belongs to Lp (0, T, W −1,p (Ω)) and u0 is in L2 (Ω). The case where f belongs to L1 (QT ) is investigated in [,1920 ] while the case w 6= 1, is studied in [,222 ] where in the first work the authors have 0 0 studied the variational case (f ∈ Lp (0, T, W −1,p (Ω, w∗ )) and in the second work the L1 (QT ) case is treated.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

125

Let us mention that in the literature of unilateral problems, the elliptic case is more studied, while the study of parabolic case is poor.

2. Preliminaries and basic assumptions Let Ω be a bounded open subset of IRN , p be a real number such that 1 < p < ∞ and w = {wi (x), 1 ≤ i ≤ N } be a vector of weight functions, i.e., every component wi (x) is a measurable function which is strictly positive a.e. in Ω. Further, we suppose in all our considerations that, there exists r0

r0 > max(N, p) such that wir0 −p ∈ L1loc (Ω),

(1)

and −1

wip−1 ∈ L1loc (Ω),

(2)

for any 0 ≤ i ≤ N . We denote by W 1,p (Ω, w) the space of all real-valued functions u ∈ Lp (Ω, w0 ) such that the derivatives in the sense of distributions fulfill ∂u ∈ Lp (Ω, wi ) for all i = 1, ..., N. ∂xi Which is a Banach space under the norm, kuk1,p,w =

"Z

Ω

|u(x)|p w0 dx +

N Z X i=1

∂u(x) p | wi (x) dx | ∂xi Ω

# p1

.

(3)

The condition (1) implies that C0∞ (Ω) is a subset of W 1,p (Ω, w) and consequently, we can introduce the subspace W01,p (Ω, w) of W 1,p (Ω, w) as the closure of C0∞ (Ω) with respect to the norm (3). Moreover, the condition (2) implies that W 1,p (Ω, w) as well as W01,p (Ω, w) are reflexive Banach spaces. We recall that the dual space of weighted Sobolev spaces W01,p (Ω, w) is 0 0 equivalent to W −1,p (Ω, w∗ ), where w∗ = {wi∗ = wi1−p , i = 0, ..., N } and p where p0 is the conjugate of p, i.e., p0 = p−1 . For more details, we refer the 12 reader to. Now we state the following assumptions: Assumption (H1 ) For 2 ≤ p < ∞, we suppose that the expression k|uk| =

N Z X i=1

∂u p | wi (x) dx | Ω ∂xi

! p1

(4)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

126

is a norm on W01,p (Ω, w) which is equivalent to (3) and that there exists a weight function σ on Ω such that, σ ∈ L1 (Ω) and σ −1 ∈ L1 (Ω).

(5)

We assume also the Hardy inequality, Z

Ω

|u(x)|p σ dx

p1

≤c

N Z X i=1

∂u p | wi (x) dx | Ω ∂xi

! p1

,

(6)

holds for every u ∈ W01,p (Ω, w) with a constant c > 0 independent of u. Moreover, the imbedding W01,p (Ω, w) ,→ Lp (Ω, σ)

(7)

expressed by the inequality (6) is compact. Note that (W01,p (Ω, w), k|.k|) is a uniformly convex ( and thus reflexive) Banach space. N Remark 2.1. Assume that w (x) ≡ 1 and there exists ν ∈ , +∞ ∩ 0 P h h 1 P −1 , +∞ such that N

wiN −1 , wi−ν ∈ L1 (Ω) for all i = 1, ..., N.

(8)

Note that the assumptions (1) and (8) imply that, k|uk| =

N Z X i=1

∂u p | wi (x) dx | Ω ∂xi

! p1

(9)

is a norm defined on W01,p (Ω, w) and its equivalent to (3) and that, the imbedding W01,p (Ω, w) ,→,→ Lp (Ω)

(10)

is compact [ see,12 pp 46]. Thus the hypotheses (H1 ) is satisfied for σ ≡ 1. Assumption (H2 ) a(x, t, s, ξ).ξ ≥ α

N X i=1

0

wi |ξi |p ,

(11)

where c1 (x, t) is a positive function in Lp (Q), and α, β are strictly positive constants.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

127

We recall that, for k > 1 and s in IR, the truncation is defined as ( s if |s| ≤ k Tk (s) = s k |s| if |s| > k. 3. Some technical lemmas 3.1. Some functional properties of time-regularization of a function u In order to deal with time derivative, we introduce a time mollification of a function u belonging in some weighted Lebesgue space. Thus we define for all µ ≥ 0 and all (x, t) ∈ QT , Z t uµ = µ u ˜(x, s)exp(µ(s − t)) ds where u ˜(x, s) = u(x, s)χ(0,T ) (s). −∞

Proposition 3.1. ∂u

1) If u ∈ Lp (QT , wi ), then, uµ is measurable in QT , ∂tµ = µ(u − uµ ) and Z Z p1 p1 p p ≤ , |uµ | wi (x) dx dt |u| wi (x) dx dt Q

Q

i.e., kuµ kLp (QT ,wi ) ≤ kukLp(QT ,wi ) .

2) If u ∈ W01,p (QT , w), then uµ → u in W01,p (QT , w) as µ → +∞. 3) If un → u in W01,p (QT , w), then (un )µ → uµ in W01,p (QT , w). 3.2. Some weighted imbedding and compactness results In this section, we establish some imbedding and compactness results in weighted Sobolev Spaces which allow in particular to extend in the settings of weighted Sobolev spaces, some trace results and the Aubin’s and Simon’s results21 . 0 Let V = W01,p (Ω, w), H = L2 (Ω, σ) and let V ∗ = W −1,p (Ω, w∗ ), with (2 ≤ p < ∞). 0 Let X = Lp (0, T, V ). The dual space of X is X ∗ = Lp (0, T, V ∗ ) where 1 1 1 0 ∗ p0 + p = 1 and denoting the space Wp (0, T, V, H) = {v ∈ X : v ∈ X } endowed with the norm kukwp1 = kukX + ku0 kX ∗ ,

(12)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

128

which is a Banach space. Here u0 stands for the generalized derivative of u, i.e., Z T Z T u(t)ϕ0 (t) dt for all ϕ ∈ C0∞ (0, T ). u0 (t)ϕ(t) dt = − 0

0

Lemma 3.1. The Banach space H is an Hilbert space and its dual H 0 can be identified with him self, i.e., H 0 ' H. Indeed, let F : H × H → ZIR f gσ dx. (f, g) 7→ Ω

Remark that F is a symmetric bilinear form, which is also continuous and defined positively, since 21 Z 21 Z Z Z 1 1 |g|2 σ dx |f |2 σ dx f σ 2 gσ 2 dx ≤ . f gσ dx = Ω

Ω

Ω

Ω

Then, the Banach space H is an Hilbert space. Finally by a standard argument, we can identified H with its dual H 0 i.e., H 0 ' H. Lemma 3.2.2 The evolution triple V ⊆ H ⊆ V ∗ is verified. Lemma 3.3.2 Assume that, ∂un = hn + kn in D0 (Ω), ∂t 0

0

where hn and kn are bounded respectively in Lp (0, T, W 1,p (Ω, w∗ ) and in L1 (QT ) . If un is bounded in Lp (0, T, W01,p (Ω, w)), then un → u in Lploc (QT , σ). Lemma 3.4.2 Let g ∈ Lr (QT , γ) and let gn ∈ Lr (QT , γ), with kgn kLr (QT ,γ) ≤ c, 1 < r < ∞. If gn (x) → g(x) a.e in QT , then gn * g in Lr (QT , γ), where * denotes weak convergence and γ is a weight function on QT .

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

129

Lemma 3.5.2 Assume that (H1 ) and (H2 ) are satisfied and let (un ) be a sequence in Lp (0, T, W01,p (Ω, w)) such that un * u weakly in Lp (0, T, W01,p (Ω, w)) and Z [a(x, t, un , ∇un ) − a(x, t, un , ∇u)][∇un − ∇u] dxdt → 0. (13) Q

Then, un → u in Lp (0, T, W01,p (Ω, w)). Lemma 3.6.23 Let V ⊆ H ⊆ V ∗ be an evolution triple. Then the imbedding Wp1 (0, T, V, H) ,→ C([0, T ]), H) is continuous . 3.3. Main results Theorem 3.1. Let u0 ∈ L1 (Ω) such that u0 ≥ 0. Assume that (H1 ) and (H2 ) hold true. Then there exists at last one solution u ∈ C([0, T ]; L1 (Ω)) such that u(x, 0) = u0 a.e. and for all τ ∈]0, T ],   Tk (u) ∈ Lp (0, T, W01,p (Ω, w)), u ≥ ψ a.e. in Ω  Z   ∂ϕ   Sk (u(τ ) − ϕ(τ )) dx + h , Tk (u − ϕ)iQτ    ΩZ ∂t +         

Qτ

a(x, t, u, ∇u)∇Tk (u − ϕ) dx dt Z Z ≤ f Tk (u − ϕ) dx dt + Sk (u0 − ϕ(x, 0)) dx Qτ

Ω

∀ k > 0 and ∀ ϕ ∈ Kψ ∩L∞ (Q) such that where Qτ = Ω×]0, τ [.

∂ϕ ∂t

0

0

∈ Lp (0, T, W −1,p (Ω, w∗ )),

Proof. Step 1: A priori estimates Consider the approximate problem ( ∂u 1 − 1 ∂t − div(a(x, t, u , ∇u )) − T (u − ψ) = f (P ) u ∈ Lp (0, T, W01,p (Ω, w)), u (x, 0) = u0 where f → f strongly L1 (Q), u0 → u0 strongly L1 (Ω). Thanks to,2 there exists at least one solution of the problem (P ).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

130

By choosing Tγ (u − Tβ (u )), β ≥ kψk∞ as test function in (P ), we get Z , T (u − T (u ))i + h ∂u a(x, t, u , ∇u )∇u dx dt γ β ∂t β≤|u |≤β+γ Z Z 1 T 1 (u − ψ)− Tγ (u − Tβ (u )) dx dt = − f Tγ (u − Tβ (u )) dx dt Q Q (14) On the one hand, we have Z Z ∂u Sγβ (u0 ) dx (15) Sγβ (u (T )) dx − , Tγ (u − Tβ (u ))i = h ∂t Ω Ω Z s β where Sγ (s) = Tγ (t − Tβ (t)) dt, and by using the fact that 0 Z Z β Sγ (u (T )) dx ≥ 0 and | Sγβ (u0 ) dx| ≤ γku0 k, we get Ω

Ω

α

Z

β≤|u |≤β+γ i=1 Z

1 − so that

N X ∂u p ∂xi wi (x) dx dt

Q

−

Z

(16)

T 1 (u − ψ)− Tγ (u − Tβ (u )) dx dt ≤ cγ, ∀ > 0 Z

Q

Tγ (u − Tβ (u )) 1 T 1 (u − ψ)− dx dt ≤ c γ

1 Tγ (u − Tβ (u )) T 1 (u − ψ)− dx dt ≥ 0, for every β ≥ kψk∞ , γ Q we deduce by Fatou’s lemma as γ → 0 that Z 1 (17) T 1 (u − ψ)− ≤ c. Q since −

Using in (P ) the test function T (u )χ(0,τ ) , we get for every τ ∈ (0, T ), Z Z Sk (u (τ )) dx + a(x, t, Tk (u ), ∇Tk (u ))∇Tk (u ) dx dt Ω Z Qτ T 1 ((u − ψ)− )Tk (u ) dx dt ≤ ck − 1 Q

which gives thanks to (17) Z Z Sk (u (τ )) dx + Ω

Qτ

a(x, t, Tk (u ), ∇Tk (u )) dx dt ≤ ck.

(18)

Then,

Z X N ∂Tk (u ) p α ∂xi wi (x) dx dt ≤ ck, ∀ k ≥ 1. Q i=1

(19)

December 29, 2009

9:37

WSPC - Proceedings Trim Size: 9in x 6in

recent

131

Hence, Tk (u ) is bounded in Lp (0, T, W01,p (Ω, w)). Let k > 0 large enough and BR be a ball of Ω, we have, Z TZ |Tk (u )| dx dt k meas({|u | > k} ∩ BR × [0, T ]) = 0 {|u |>k}∩BR Z TZ ≤ |Tk (u )| dx dt (20) 0 BR ! Z p1 Z TZ 0 ≤ × |Tk (u )|p w0 dx dt w01−p dx dt Q

0

BR

then, thanks to (H1 ), we deduce that, ! p1 Z X N ∂Tk (u ) p k meas({|u | > k} ∩ BR × [0, T ]) ≤ c ∂xi wi (x) dx dt Q i=1 1

≤ ck p

(21)

which implies that, meas({|u | > k} ∩ BR × [0, T ]) ≤

k

c1 1 1− p

,

∀ k ≥ 1.

So, we have, lim (meas({(x, t) ∈ Q : |u | > k} ∩ BR × [0, T ]) = 0

k→+∞

(22)

uniformly with respect to . Consider now a function nondecreasing ξk ∈ C 2 (IR) such that ξk (s) = s for |s| ≤ k2 ξk (s) = k for |s| ≥ k. Multiplying the approximate equation by ξk0 (u ), we get ∂ ∂t (ξk (u ))

− div(a(x, t, u , ∇u )ξk0 (u )) + a(x, t, u , ∇u )ξk00 (u ) − 1 T 1 ((u − ψ)− )ξk0 (u ) = f ξk0 (u ),

in the sense of distribution. This implies, thanks to (19) and the fact that ξk0 has compact support, that ξk (u ) is bounded in Lp (0, T, W01,p (Ω, w)), while it’s time derivative ∂ p0 −1,p0 (Ω, w∗ )) + L1 (QT ), hence lemma ∂t (ξk (u )) is bounded in L (0, T, W 3.3 allows us to conclude that ξk (u ) is compact in Lploc (QT , σ). Thus, for a subsequence, it also converges in measure and almost every where in QT since we have, for every λ > 0

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

132

meas({|u − uη | > λ} ∩ BR × [0, T ]) ≤ meas({|u | > k2 } ∩ BR × [0, T ]) +meas({|uη | > k2 } ∩ BR × [0, T ]) +meas({|ξk (u ) − ξk (uη )| > λ} ∩ BR × [0, T ]). (23) p Let σ > 0, then, by (22) and the fact that ξk (u ) is compact in Lloc (QT , σ), there exists k(σ) > 0 such that, meas({|u − uη | > λ} ∩ BR × [0, T ]) ≤ σ for all , η ≤ 0 (k(σ), λ, R). This proves that (u ) is a Cauchy sequence in measure in BR × [0, T ], thus converges almost everywhere to some measurable function u. Then for a subsequence denoted again u , we can deduce from (19) that, Tk (u ) * Tk (u) weakly in Lp (0, T, W01,p (Ω, w)).

(24)

and then, the compact imbedding (7) gives, Tk (u ) → Tk (u) strongly in Lp (QT , σ) and a.e. in QT .

(25)

Step 2: About the gradient of approximate solutions In the sequel and throughout the paper, we will denote α(, µ, s) all quantities (possibly different) such that, lim lim lim α(, µ, s) = 0.

s→∞ µ→∞ →+0

Taking now Tη (u − (Tk (u))µ ), η > 0 as test function in (P ), we get Z h ∂u , T (u − (T (u)) )i + a(x, t, u , ∇u )∇Tη (u − (Tk (u))µ ) η k µ ∂t Q Z − 1 T 1 ((u − ψ)− )Tη (u − (Tk (u))µ ) dx dt ≤ cη, QT

which implies that,

Z a(x, t, u , ∇u )∇Tη (u − (Tk (u))µ ) h ∂u , T (u − (T (u)) )i + η k µ ∂t QT Z ≤ η T 1 ((u − ψ)− ) dx dt + cη QT

and by (17) h ∂u ∂t , Tη (u − Tk (u)µ )i +

Z

QT

a(x, t, u , ∇u )∇Tη (u − (Tk (u)µ )

(26)

≤ cη.

The first term of the left-hand side of the last inequality reads as, ∂T (u)

µ ∂u k h ∂u , Tη (u − Tk (u)µ )i ∂t , Tη (u − Tk (u)µ )i = h ∂t − ∂t ∂Tk (u)µ +h ∂t , Tη (u − Tk (u)µ )i.

(27)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

133

The second term of the last equality can be written as, Z ∂Tk (u)µ h ∂u − , T (u − T (u) )i = Sη (u (T ) − Tk (u)µ (T )) dx η k µ ∂t ∂t Ω Z Z − Sη (u0 ) dx ≥ −η |u0 | dx Ω

(28)

Ω

≥ −ηc.

The third term can be written as, Z ∂Tk (u)µ (Tk (u)−Tk (u)µ )(Tη (u −Tk (u)µ )) dx dt h , Tη (u −Tk (u)µ )i = µ ∂t QT (29) thus by letting → 0 and by using Lebesgue theorem, Z (Tk (u) − Tk (u)µ )(Tη (u − Tk (u)µ )) dx dt QZ T = (Tk (u) − Tk (u)µ )(Tη (u − Tk (u)µ )) dx dt. QT

Consequently, h

∂u , Tη (u − Tk (u)µ )i ≥ α(, µ) − ηc ∂t

(30)

on the other hand, Z a(x, t, u , ∇u )∇Tη (u − Tk (u)µ ) dx dt QZ T = a(x, t, u , ∇u )(∇u − ∇Tk (u)µ ) dx dt Z{|u −Tk (u)µ )|<η} = a(x, t, Tk (u ), ∇Tk (u ))(∇Tk (u ) − ∇Tk (u)µ ) dx dt {|T Zk (u )−Tk (u)µ )|<η} + a(x, t, u , ∇u )(∇u − ∇Tk (u)µ ) dx dt {|u |>k}∩{|u −Tk (u)µ )|<η}

which implies, by using the fact that Z a(x, t, u , ∇u )∇u dx dt ≥ 0, {|u |>k}∩{|u −Tk (u)µ )|<η}

that Z

{|Tk (u Z )−Tk (u)µ )|<η}

≤ cη +

a(x, t, Tk (u ), ∇Tk (u ))(∇Tk (u ) − ∇Tk (u)µ ) dx dt

{|u |>k}∩{|u −Tk (u)µ )|<η}

a(x, t, u , ∇u )|∇Tk (u)µ | dx dt. (31)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

134

Since a(x, t, Tk+η (u ), ∇Tk+η (u )) is bounded some hk+η ∈

N Y

N Y

0

Lp (QT , wi∗ ), there exists

i=1

0

Lp (QT , wi∗ ) such that,

i=1

a(x, t, Tk+η (u ), ∇Tk+η (u )) * hk+η weakly in Consequently, Z

{|uZ |>k}∩{|u −Tk (u)µ )|<η}

N Y

0

Lp (QT , wi∗ ).

i=1

a(x, t, u , ∇u )|∇Tk (u)µ | dx dt

=

{|u|>k}∩{|u−Tk (u)µ )|<η}

hk+η |∇Tk (u)µ | dx dt + α()

thanks to proposition 3.1, one easily has, Z hk+η |∇Tk (u)µ | dx dt + α() = α(, µ). {|u|>k}∩{|u−Tk (u)µ )|<η}

Hence, Z

{|Tk (u )−Tk (u)µ )|<η}

a(x, t, Tk (u ), ∇Tk (u ))(∇Tk (u ) − ∇Tk (u)µ ) dx dt

≤ cη + α(, µ).

(32) On the other hand, note that Z a(x, t, Tk (u ), ∇Tk (u ))(∇Tk (u ) − ∇Tk (u)µ ) dx dt {|T Z k (u )−Tk (u)µ )|<η} = a(x, t, Tk (u ), ∇Tk (u ))(∇Tk (u ) − ∇Tk (u)) dx dt {|T Z k (u )−Tk (u)µ )|<η} + a(x, t, Tk (u ), ∇Tk (u ))(∇Tk (u) − ∇Tk (u)µ ) dx dt {|Tk (u )−Tk (u)µ )|<η}

(33)

the last integral tends to 0 as → 0 and µ → ∞. Indeed, we have that Z a(x, t, Tk (u ), ∇Tk (u ))(∇Tk (u) − ∇Tk (u)µ ) dx dt → Z{|Tk (u )−Tk (u)µ )|<η} hk (∇Tk (u) − ∇Tk (u)µ ) dx dt as → 0. {|Tk (u)−Tk (u)µ )|<η}

It is obviously that, Z

{|Tk (u)−Tk (u)µ )|<η}

hk (∇Tk (u) − ∇Tk (u)µ ) dx dt → 0 as µ → ∞.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

135

We deduce then that, Z

{|Tk (u )−Tk (u)µ )|<η}

a(x, t, Tk (u ), ∇Tk (u ))(∇Tk (u ) − ∇Tk (u)) dx dt

≤ cη + α(, µ).

(34)

Let A be expression in brace above, then for any 0 < η < 1 Z Z Aθ dx dt Aθ dx dt + I = {|Tk (u )−Tk (u)µ )|>η}

{|Tk (u )−Tk (u)µ )|≤η}

since, a(x, Tk (u ), ∇Tk (u )) is bounded in ∇Tk (u )L∞ bounded in equality, we obtain, Z I ≤ c

N Y

N Y

0

0

Lp (QT , wi1−p ), while

i=1

Lp (QT , wi ), then by applying the H¨ older’s in-

i=1

A dx dt

{|Tk (u )−Tk (u)µ )|<η}

!θ

+c2 meas{(x, t) ∈ QT : |Tk (u ) − Tk (u)µ )| > η}

(35) 1−θ

,

on the other hand, we have, Z A dx dt {|T Z k (u )−Tk (u)µ )|<η} = a(x, t, Tk (u ), ∇Tk (u ))(∇Tk (u ) − ∇Tk (u)) dx dt Z {|Tk (u )−Tk (u)µ )|<η} − a(x, t, Tk (u ), ∇Tk (u))(∇Tk (u ) − ∇Tk (u)) dx dt {|Tk (u )−Tk (u)µ )|<η}

= I1 + I2

(36) using (34), we have, I1 ≤ cη + α(, µ).

(37)

I2 = α()

(38)

Concerning I2 the second term of the right hand side of the (36), it is easy to see that

because for all i = 1, . . . , N , we have, ai (x, t, Tk (u ), ∇Tk (u)) → 0 0 k (u ) k (u) ai (x, t, Tk (u), ∇Tk (u)) strongly in Lp (QT , wi1−p ), while ∂T∂x * ∂T∂x i i weakly in Lp (QT , wi ). Combining (35), (36), (37) and (38) we get, I ≤ c meas{|Tk (u ) − Tk (u)µ )| < η}θ + c(α(, µ, η))1−θ

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

136

and by passing to the limit sup over , µ and η Z [a(x, t, Tk (u ), ∇Tk (u )) lim →0

QT

−a(x, t, Tk (u ), ∇Tk (u))][∇Tk (u ) − ∇Tk (u)]

θ

(39) dx dt = 0

and thus there exist a subsequence also denoted by u such that, ∇u −→ ∇u a.e. in QT .

(40)

Step 3: Passage to the limit ¯ choosing Tk (u − ϕ)χ Let ϕ ∈ Kψ ∩ L∞ (Q), as test function in (P ), we (0,τ ) get, Z h ∂u a(x, t, u , ∇u )∇Tk (u − ϕ) dx dt , T (u − ϕ)i + k Qτ ∂t Qτ Z T 1 (u − ψ)− T (u − ϕ) dx dt − 1 (41) Z Qτ f T (u − ϕ) dx dt = Qτ

since, − 1 we get

Z

Qτ

T 1 (u − ψ)− T (u − ϕ) dx dt ≥ 0 and

∂u ∂t

=

∂ϕ ∂ ∂t (u − ϕ) + ∂t ,

Z

∂ϕ Sk (u (τ ) − ϕ(τ )) dx + h , Tk (u − ϕ)iQτ ∂t ΩZ a(x, t, u , ∇u )∇Tk (u − ϕ) dx dt + Z ZQτ Sk (u (0) − ϕ(0)) dx. f T (u − ϕ) dx dt + ≤ Qτ

(42)

Ω

Lemma 3.7. The sequence (u ) is a Cauchy sequence in C([0, T ], L1 (Ω)), moreover, u ∈ C([0, T ], L1 (Ω)) and (u ) converges to u in C([0, T ], L1 (Ω)). This lemma will be proved below. Because of u → u in C([0, T ], L1 (Ω)), then ∀ τ ≤ T , u (t) → u(t) in L1 (Ω), thus Z Z Sk (u (τ ) − ϕ(τ )) dx → Sk (u − ϕ) dx (43) Ω

and

Z

Ω

Sk (u (0) − ϕ(0)) dx →

Ω

Z

Ω

Sk (u0 − ϕ(0)) dx.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

137

Let M = k + kϕk∞ , then, we can write, Z a(x, t, un , ∇un )∇Tk (un − ϕ) dx dt QZ T = a(x, t, TM (un ), ∇TM (un ))∇Tk (un − ϕ) dx dt, QT

which implies by Fatou’s lemma (and the fact a(x, t, TM (un ), ∇TM (un )) * a(x, t, TM (u), ∇TM (u)) weakly in

N Y

0

0

Lp (Q, wi1−p ), that we can deduce,

i=1

Z

QT

a(x, t, TM (u), ∇TM (u))∇Tk (u − ϕ) Z a(x, t, TM (u ), ∇TM (u ))∇Tk (u − ϕ) dx dt. ≤ lim inf →0

(44)

Ω

0

0

p −1,p (Ω, w∗ )) and ∇Tk (u −ϕ) * ∇Tk (u− Moreover, since ∂ϕ ∂t ∈ L (0, T, W N Y ϕ) weakly in Lp (QT , wi ), we get, i=1

Z

Qτ

Z

∂ϕ Tk (u − ϕ) dx dt → ∂t

Z

f Tk (u − ϕ) dx dt →

Z

Q

Qτ

Q

∂ϕ Tk (u − ϕ) dx dt ∂t

f Tk (u − ϕ) dx dt.

(45)

(46)

Finally, by (42)-(46) we get, Z ∂ϕ Sk (u (τ ) − ϕ(τ )) dx + h , Tk (u − ϕ)iQτ ∂t Ω Z + a(x, t, u, ∇u)∇Tk (u − ϕ) dx dt Qτ Z Z ≤ f Tk (u − ϕ) dx dt + Sk (u(0) − ϕ(0)) dx, Qτ

Ω

which completes the proof of the theorem. Proof of Lemma 3.7: Since Tl (u) ∈ Kψ , for every l ≥ kψk∞ . Let vµi,l = (Tl (u))µ + e−µt Tl (ηi ) with ηi ≥ 0 converges to u0 in L1 (Ω), as

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

138

test function in (46), h ∂u ∂t , Tk (u

−

vµi,l )iQτ

Z

+ Z

− 1 Z =

Qτ

Qτ

Qτ

a(x, t, u , ∇u )∇Tk (u − vµi,l ) dx dt

T 1 (u − ψ)− Tk (u − vµi,l ) dx dt

(47)

f Tk (u − vµi,l ) dx dt

since, ∂ ∂ ∂u = (u − vµi,l ) + (vµi,l ) ∂t ∂t ∂t ∂ = (u − vµi,l ) + µ(Tl (u) − vµi,l ) ∂t we deduce, h

∂u ∂ , Tk (u − vµi,l )iQτ = h (u − vµi,l ), Tk (u − vµi,l )i + α(, µ, l) ∂t ∂t

which implies that, h

∂u ∂ (u − vµi,l ), Tk (u − vµi,l )i = h , Tk (u − vµi,l )iQτ + α(, µ, l) ∂t ∂t

(48)

onZthe other hand, by using the monotonicity of a and the fact that, 1 − T 1 (u − ψ)− Tk (u − vµi,l ) dx dt ≥ 0, we deduce that by (47), Qτ Z i,l + h ∂u , T (u − v )i a(x, t, u , ∇vµi,l )∇Tk (u − vµi,l ) dx dt k Qτ µ ∂t QτZ ≤ f Tk (u − vµi,l ) dx dt Qτ

since for all i = 1, ..., N ai (x, t, T2k+l (u ), ∇vµi,l ) → ai (x, t, Tk+2l (u), ∇vµi,l ) 0

0

strongly in Lp (Q, wi1−p ) while in Lp (Q, wi ), we have, h

∂ ∂xi Tk (u

− vµi,l ) *

∂ i,l ∂xi Tk (u − vµ )

∂u , Tk (u − vµi,l )i ≤ α(, µ, i, l). ∂t

weakly

(49)

Therefore, by writing Z ∂vµi,l ∂u Sk (u (τ ) − vµi,l (τ )) dx = h , Tk (u − vµi,l )iQτ − h , Tk (u − vµi,l )iQτ ∂t ∂t Ω Z +

Ω

Sk (u0 − Tl (ηi )) dx

(50)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

139

and using (48) and (49) we get, Z Sk (u (τ ) − vµi,l (τ )) dx ≤ α(, µ, i, l)

(51)

Ω

which implies, by writing, Z Z u − u λ 2 Sk ( ) dx ≤ Sk (u (τ ) − vµi,l (τ )) + Sk (uλ (τ ) − vµi,l (τ )) dx, 2 Ω Ω that

Z

Sk ( Ω

u − u λ ) dx ≤ α(, λ). 2

(52)

Finally, by H¨ older’s inequality, we have, Z Z Z |u − uλ | dx |u − uλ | dx + |u − uλ | dx = {|u −uλ |>1} {|u −uλ |≤1} Ω 1 ! Z Z 2 ≤

{|u −uλ |≤1}

≤ meas(Ω)

1 2

Z

|u − uλ |2 dx

1

meas(Ω) 2 + 2

{|u −uλ |≤1}

2S1 |u − uλ | dx

! 21

{|u −uλ |>1}

+

Z

|u − uλ | dx

{|u −uλ |>1}

2S1 (u − uλ ) dx

since ( 2

|y| |y| |y| − 1 )χ ≤( + )χ{|y|>1} = S1 (y)χ{|y|>1} 2 {|y|>1} 2 2

and ( |y|2 )χ{|y|≤1} = S1 (y)χ{|y|≤1} then by (52) we deduce that,

Z

Ω

|u (τ ) −

uλ (τ )| dx ≤ α(, λ), not depending on τ . And thus (u ) is a Cauchy sequence in C([0, T ], L1(Ω)), and since u → u, a.e. in Q, we deduce that u → u, a.e. in C([0, T ], L1 (Ω)). References 1. R. Adams, Sobolev spaces, AC, Press, New York, (1975) 2. L. Aharouch, E. Azroul and M. Rhoudaf Strongly nonlinear variational parabolic problems in Weighted Sobolev spaces to appear in AJMAA. 3. Y. Akdim, E. Azroul and A. Benkirane, Existence Results for Quasilinear Degenerated Equations Via Strong Convergence of Truncations, Revista Matematica Complutense 17, , N.2, (2004) pp 359-379. 4. Y. Akdim, E. Azroul and A. Benkirane, Existence of Solution for quasilinear degenerated Elliptic Unilateral Problems, Ann. Math. Blaise pascal vol 10 (2003) pp 1-20. 5. Y. Akdim, E. Azroul and A. Benkirane, Existence of solution for quasilinear degenerated elliptic equation, Electronic J. Diff. Equ. Vol 2001 , N71, (2001) pp 1-19.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

140

6. H. Brezis, and F.E. Browder, Strongly nonlinear parabolic initial-boundary value problems , Proc. Nat Acad. Sci. U. S. A. 76 (1976). pp. 38-40. 7. J. Berkovits, V. Mustonen, Topological degree for perturbation of linear maximal monotone mappings and applications to a class of parabolic problems,Rend.Mat.Roma,Ser,VII, 12 (1992), pp. 597-621. 8. L. Boccardo, F. Murat, Strongly nonlinear Cauchy problems with gradient dependt lower order nonlinearity, Pitman Research Notes in Mathematics, 208 (1988), pp. 347-364. 9. L. Boccardo, F. Murat, Almost everywhere convergence of the gradients of solutions to elliptic and parabolic equations, Nonlinear analysis, T.M.A., 19 (1992), n 6, pp. 581-597. 10. A. Dallaglio A. Orsina , Non linear parabolic equations with natural growth condition and L1 data. Nolinear Anal., T.M.A., 27 n1 (1996). pp. 59-73. 11. P. Drabek, A. Kufner and L. Mustonen, Pseudo-monotonicity and degenerated or singular elliptic operators, Bull. Austral. Math. Soc. Vol. 58 (1998), 213-221. 12. P. Drabek, A. Kufner and F. Nicolosi, Non linear elliptic equations, singular and degenerated cases, University of West Bohemia, (1996). 13. A. Kufner, Weighted Sobolev Spaces, John Wiley and Sons, (1985). 14. R. Landes, On the existence of weak solutions for quasilinear parabolic initial-boundary value problems, Proc. Roy. Soc. Edinburgh sect. A. 89 (1981), 217-137. 15. R. Landes, V. Mustonen, A strongly nonlinear parabolic initial-boundary value problems, Ark. f. Math. 25. (1987). 16. R. Landes, V. Mustonen, On parabolic initial-boundary value problems with critical growth for the gradient, Ann. Inst. H. Poincar´e11 (2) (1994) 135-158. ˙ sik ˙ sur les probl`emes elliptiques 17. J. Leray, J.L. Lions, Quelques resultats de Viˇ nonlin´eaires par les m´ethodes de Minty-Browder, Bull. Soc. Math. France 93 (1995), 97-107. 18. J.L. Lions, quelques methodes de r´esolution des probl`emes aux limites non lin´eaires, Dunod et Gauthiers-Villars, 1969. 19. A. Porretta Existence results for nonlinear parabilc equations via strong convergence of truncations, Ann. Mat. Pura. Appl. (1999), pp. 143-172. 20. J. M. Rakotoson A Compactness lemma for quasilinear problems: application to parabolic equations J. Funct. Anal. 106 (1992), pp. 358-374. 21. J. Simon Compact sets in the space Lp (0, T, B), Ann. Mat. Pura. Appl. 146 (1987), pp. 65-96. 22. M. Rhoudaf Existence results for Strongly nonlinear degenerated parabolic equations via strong convergence of truncations with L1 -data th`ese de Doctorat de M.Rhoudaf, Univ. Sidi Mohamed Ben Abdella. F`es, Maroc 2006. 23. E. Zeidler, nonlinear functional analysis and its applications, II A and II B, Springer-Verlag (New York-Heidlberg, 1990).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

141

Numerical analysis of slopes stability under seismic loading in Lebanon Fadi Hage Chehade Universit´ e Libanaise, Institut Universitaire de Technologie BP 813 Saida, Lebanon E-mail: [email protected] Marwan Sadek Laboratoire de M´ ecanique de Lille, UMR 8107 Universit´ e des Sciences et Technologies de Lille, France E-mail: [email protected] Most common methods used in engineering practice to assess the seismic stability of slope consists of a pseudo static approach or traditional approaches based on the limit equilibrium or on the hypothesis of failure calculation. Such methods are widely used because of their simplicity and there is no need of sophisticated software for their application. However, they neglect important elements such as the soil deformability, the dynamic amplification and non linear soil behavior. The present study concerns the analysis of the seismic slopes stability using global dynamic approach. It will mainly focus on dynamic amplification in the slope under real earthquake records with frequency content close to the natural frequency of the slope. Analysis is conducted by numerical modeling using Flac3D finite difference program. Results are presented in the case of linear visco-elastic then elastoplastic behaviour for earthfill materials in order to elucidate the influence of plasticity. Keywords: Seismic slope; Numerical modeling; Elastoplastic behavior.

1. Introduction Recent devastating earthquakes in Pakistan, Turkey, Algeria and recently in China call to the mind the high risk exposure of Lebanon. This country is located over active fault (Harajli et al., 2002) and many geologists and experts shared the view that a major seismic event may occur in Lebanon in the future. Moreover, many earthquakes (more than 500 according to the National Council of Scientific Research in Lebanon), of low magnitudes between three and five, have been registered in Lebanon between February

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

142

and October 2008. These events lead to great anxiety among Lebanese population because of high risk exposure in case of earthquake. Indeed there is no sufficient protection in most of civil constructions to seismic loading. It is well known that the slopes represent a weak area where the seismic consequences could be amplified (Davidovici, 1999). Large movement such as soil collapse, bloc rocks fall, rock or soil sliding could arise in steep slopes (angle more than 35) in case of earthquake [Keefer, 1984 ; Rodriguez et al., 1999]. Lebanon topography contains a lot of slopes and actually they become highly occupied because of the population migration from urban to surrounded suburb areas. Around the capital Beirut, it can be seen several slopes zones with high density of population. The most common method used in engineering practice to assess the seismic stability of slope consists on a pseudo static approach where the earthquake effect on a potential soil mass is represented by means of equivalent static horizontal force equal to the soil mass multiplied by a seismic coefficient. This approach is quite simplistic since it attempts to represent complex dynamic behaviour in terms of static forces. Stability is expressed in terms of an overall factor of safety. The implicit assumption is that the soil is rigid-perfectly plastic behaving as an undeformable block. Other traditional approaches are based on the limit equilibrium or on the hypothesis of failure calculation (for example Bishop (1955)). They are widely used in engineering practice because of their simplicity and there is no need of sophisticated software for their application. The main assumptions are based on the soil behavior (rigid bloc) and on the failure modes and their localization (circular, plane, . . . ). Progress in the area of geotechnical computation and numerical modeling offers interesting facilities for the analysis of the seismic induced response of soil and structure systems in considering complex issues such as the soil non linearity, the evolution of the pore pressure and real earthquake records. Detailed analysis techniques include equivalent linear (decoupled) solutions, and non linear finite element and finite difference coupled or decoupled formulations (Lin and Chao 1990, Abouseeda and Dakoulas 1998, Cascone and Rampello 2003, . . . ). Wood (1973) showed that where the principal energy of the input motions approaches the fundamental frequency of the unrestrained backfill, dynamic amplification becomes an important factor, which is not considered in engineering approaches that assess the earth fill dam stability. The present paper proposes a numerical study of the seismic behaviour of slopes. It will mainly focus on dynamic amplification in the slope under earthquake

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

143

loading with frequency content close to the natural frequency of the slope. In the present study, analysis is conducted using a finite difference modeling. Results are presented in the cases of linear visco-elastic and elastoplastic behaviours for earthfill materials in order to elucidate the influence of plasticity. Indeed, since we deal with unconfined material, the seismic loading generally induces plasticity in the soil. This plasticity can influence the all over response of the slope, because of its influence on damping and on the dominant frequencies. 2. Dynamic analysis 2.1. Problem under consideration and basic equations The selected example is a simplified representation of typical slope geometry (Figure 1). Mechanical properties of the soil used in the analyses are presented in Table 1. They correspond to the case of one layer of unconsolidated weak soil. Numerical analyses are conducted using the finite difference FLAC3D program based on a continuum finite difference discretization using the Langrangian approach (Flac3D, 2005). In the Lagrangian formulation adopted in FLAC3D, a point in the dvi , medium is characterized by the vector components xi , ui , vi and dt i = 1, 3 of position, displacement, velocity and acceleration, respectively. The state of stress at a given point of the medium is characterized by the symmetric stress tensor σij . The traction vector [t] on a face with unit normal [n] is given by Cauchy’s formulae ti = σij .nj . In an infinitesimal time dt, the medium experiences an infinitesimal strain determined by the translations vi dt, and the corresponding components of the strain-rate tensor may be written as 1 (vi,j + vj,i ) 2 where partial derivatives are taken with respect to components of the current position vector [x]. Application of the continuum form of the momentum principle yields Cauchy’s equations of motion : ξij =

ρdvi dt where ρ is the mass per unit volume of the medium, [b] is the body force per d[v] unit mass, and is the material derivative of the velocity. These laws dt σij,j + ρbi =

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

144

govern, in the mathematical model, the motion of an elementary volume of the medium from the forces applied to it. FLAC3D is based on a continuum finite difference discretization using the Langrangian approach. Every derivative in the set of governing equations is replaced directly by an algebraic expression written in terms of the field variables (e.g. stress or displacement) at discrete point in space. An important aspect of the model is the inclusion of the equations of motion; the calculation sequence first invokes the equations of motion to derive new velocities and displacements from stresses and forces. Then, strain rates are derived from velocities, and new stresses from strain rates. Every cycle around the loop correspond to one time step. The final solution is reached (using a damped solution) when the body is in equilibrium or in steadystate flow (plastic flow), and the out of balance force goes to zero. Dynamic loading is applied at the base of the foundation layer as a velocity excitation. Free field boundaries were applied at the sides of the model in order to stop boundary effect. Kuhlemeyer and Lysmer (1973) showed that for an accurate representation of the wave transmission through the soil model, the spatial element size, ∆l, must be smaller than approximately one-tenth to one-eighth of the wavelength associated with the highest frequency component of the input wave i.e., ∆l ≤

λ . 10

(1)

Here, λ is the wave length associated with the highest frequency component that contains appreciable energy. The consequence is that reasonable analyses may be time and memory consuming. In such cases, it may be possible to adjust the input by recognizing that most of the power for the input history is contained in lower frequency components (< 10Hz). Rayleigh damping of 5% is used in the analysis to compensate the energy dissipation through the medium (Paolucci, 2002; Lokmer et al., 2002). When plasticity is considered, damping occurs mainly through hysteretic looping, Rayleigh damping is fixed at 2% ; the. The maximum length of element is fixed to 1m in both vertical and horizontal directions (Figure 2).

2.2. Elastic response In order to quantify the influence of frequency loading, numerical simulations has been firstly conducted under harmonic loading composed of 15

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

145

Fig. 1.

Fig. 2.

A typical slope geometry.

Slope general configuration and finite difference mesh.

cycles. Figures 2 and 3 depict the maximum lateral amplifications according to the loading frequency applied at the base. It can be noted that where the applied frequency approach the fundamental frequency of the soil mass, the lateral amplification increase significantly from the bottom to the slope crest. For example, when the frequency loading fch equal the fundamental frequency of the system, the lateral amplification at the crest (10.3) is 3 times higher than that obtained at the bottom (3.4). In this

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

146

case, it’s evident that neglecting soil deformability as used in pseudo-static approach, leads to significant error. From the other hand, when load frequency moves aside the fundamental frequency, lateral amplification shows little discrepancy between the top and the crest of the slope. In this case, the hypothesis of constant acceleration could be accepted as presumed in simplified approach. 2.3. Influence of plasticity In order to examine the influence of plasticity, the slope system has been subjected to earthquake loading representative of the 1999 Kocaeli earthquake in Turkey (M w = 7.4, Chen and Scawthorn 2003). Note that the dominant frequency of applied load is about 0.9 Hz which is close to the fundamental frequency of the system. Figure 4 shows a comparison between the elastic and elastoplastic analyses at the maximum lateral amplification. It can be observed that the plastic deformation leads to a decrease in the velocity amplification, in particular in the upper part (Figures 5 and 6). The reduction is dependent on the intensity of the applied loading (Figure 6). For example, it attains about 50% when the velocity (v) is equal to 2m/s. This result could be attributed to the energy dissipation by plastic deformation and to the influence of plasticity on the reduction of the fundamental frequencies of the slope. When the applied velocity exceeds 0.5m/s which corresponds to a moderate earthquake, the lateral amplification show little discrepancy between the bottom and the top. For lower values, lateral amplification shows important difference with the depth and the hypothesis of constant soil amplification is not valid.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

147

Fig. 3.

Amplification at different positions of the slope according to the load frequency.

Fig. 4. ing.

Variation of maximum lateral amplification along the slope with frequency load-

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

148

Fig. 5.

Fig. 6.

A decrease in the velocity amplification.

Influence of plasticity on lateral amplification.

3. Conclusion This paper included analysis of the seismic stability of slopes. Analyzes were conducted for harmonic and real earthquake records. For dynamic loading with dominant frequency close to the fundamental frequency of the slope, lateral amplification show a important discrepancy along the depth and the hypothesis of constant acceleration used in pseudo static approach is not

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

149

valid. Elastoplastic analyzes show that the plastic deformation leads to a decrease in the velocity amplification, in particular in the upper part. Parametric analysis shows for moderate earthquake, the dynamic amplification still a important parameter and should be taken into account in any stability survey. Work is under progress to incorporate the variation of dynamic amplification in simplified approaches. References 1. H. Abouseeda, and P. Dakoulas, Non-Linear Dynamic Earth DamFoundation Interaction Using a BE-FE Method, Earthquake Engineering and Structural Dynamics, Vol. 27 (1998), pp. 917-936. 2. A. Bishop, The use of the slip circle in the stability analysis of slopes, Geotechnique, 5(1) (1955), pp. 7-17. 3. E. Cascone, and S. Rampello, Decoupled Seismic Analysis of an Earth Dam, Soil Dynamics and Earthquake Engineering, Vol. 23 (2003), pp. 349-365. 4. W.F. Chen and C. Scawthorn, Earthquake Engineering Handbook, CRC Press LLC, 2003 5. V. Davidovici, la construction en zone sismique, Editions Le Moniteur, Paris (1999). 6. M. Harajli, S. Sadek and R. Aspahan, Evaluation of the seismic hazard of Lebanon, Journal of the Seismology, 6 (2002), pp. 257-277. 7. Itasca Consulting Group, FLAC: Fast Lagrangian Analysis of Continua, vol. I. User’s Manual, vol. II. Verification Problems and Example Applications, Second Edition (FLAC3D Version 3.0), Minneapolis, Minnesota 55401 USA, 2005. 8. D.K. Keefer, Landslides caused by earthquakes, Bulletin of the seismological society of America, 95 (1984), pp. 406-421. 9. R. L.Kuhlmeyer and J. Lysmer,, Finite Element Method Accuracy for Wave Propagation Problems, J. Soil Mech. and Foundations Div., ASCE, 99(SM5) (1973), pp. 421-427. 10. J. Lin and B. Chao, Estimation of Shear Moduli and Damping Factors of Earth Dam Materials, Earthquake Engineering and Structural Dynamics, Vol. 19 (1990), pp. 891-910. 11. R. Paolucci, Amplification of earthquake ground motion by steep topographic irregularities”, Earthquake Engineering and Structural dynamics, 2002; 31, pp.1831-1853. 12. C.E. Rodriguez, J. J. Bommer and R. J. Chandler, Earthquake-induced landslides: 1980-1997. Soil dynamics earthquake engineering, 18 (1999), pp. 325346. 13. J. Wood, Earthquake-Induced Soil Pressures on Structures,” Report EERL 73-05, California Institute of Technology, Pasadena 1973.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

150

Existence and uniqueness result for a class of nonlinear parabolic equations with L1 data Hicham Redwane Facult´ e des Sciences Juridiques, Economiques et Sociales Universit´ e Hassan 1, B.P. 764, Settat, Morocco E-mail: redwane [email protected] In this paper, we consider the initial-boundary value problem of a nonlinear parabolic equation with the data belong to L1 and no growth assumption is made on the nonlinearities. We establish the existence and partial uniqueness theorems of renormalized solutions. Keywords: Nonlinear parabolic equation; Renormalized solution; Existence and uniqueness.

1. Introduction This paper is concerned with the following initial-boundary value problem ∂b(x, u) − div A(x, t)Du + Φ(u) + f (x, t, u) = 0 in Ω × (0, T ), (1) ∂t b(x, u)(t = 0) = b(x, u0 ) in Ω,

(2)

u = 0 on ∂Ω × (0, T ),

(3)

where Ω is a bounded open set of IRN , (N ≥ 1), T is a positive real number while the data b(x, u0 ) in L1 (Ω). The matrix A(x, t) is a bounded symmetric and coercive matrix. The function b(x, s) is assumed to strictly increasing and C 1 for s (for every x ∈ Ω) but which is not restricted by any growth condition with respect to s (see assumptions (4) and (5) of Section 2), f is a Carath´eodory function in Ω × (0, T ) × IR and not controlled with respect to s. The function Φ is just assumed to be continuous on IR. A large number of papers was devoted to the study of the existence and uniqueness for solution of parabolic problems under various assumptions and in different contexts: for a review on classical results (see, e.g. ,2 ,4 ,6 ,7 ,10 ,11 12 ).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

151

When Problems (1)-(3) is investigated of difficulty is due to the facts that the data f and b(x, u0 ) only belong to L1 , the function f (x, t, u), Φ(u), does not belong L1loc (Ω × (0, T )) in general, and the main difficulty relies on the dependence of b(x, u) on both x and u, so that proving existence of a weak solution (i.e. in the distribution meaning) seems to be an arduous task. The existence of renormalized solutions (1)-(3) has been proved in H. Redwane22 in the case where f (x, t, u) is independent of u and where −div(A(x, t)Du) is replaced by the operator −div(a(x, t, u, Du)) (the operator is a Leray-Lions which is coercive and which grows like |Du|p−1 with respect to Du, but which is not restrict by any growth condition with respect to u), and the uniqueness of renormalized solutions has been proved in H. Redwane23 in the case where f (x, t, u) is independent of u. 2. Assumptions on the data and definition of a renormalized solution We take Ω a bounded open set on IRN (N ≥ 1), T > 0 is given and we set Q = Ω × (0, T ). b : Ω × IR → IR is a Carath´eodory function such that ;

(4)

for every x ∈ Ω : b(x, s) is a strictly increasing C 1 -function, with b(x, 0) = 0. For any K > 0, there exists λK > 0, a function AK in L∞ (Ω) and a function BK in L2 (Ω) such that ∂b(x, s) ∂b(x, s) λK ≤ ≤ AK (x) and ∇x (5) ≤ BK (x) ∂s ∂s for almost every x ∈ Ω, for every s such that |s| ≤ K. A(x, t) is a symmetric coercive matrix field with coefficients

(6)

lying in L∞ (Q) i.e. A(x, t) = (aij (x, t))1≤i, j≤N with aij (x, t) ∈ L∞ (Q) and aij (x, t) = aji (x, t) a.e. in Q, ∀i, j, and there exists α > 0 such that A(x, t)ξ.ξ ≥ α|ξ|2 a.e. in Q, ∀ξ ∈ IRN . Φ : IR → IRN is a continuous function. f : Q × IR → IR

is a Carath´eodory function.

(7) (8)

For almost every (x, t) ∈ Q, for every s ∈ IR: sign(s)f (x, t, s) ≥ 0 and f (x, t, 0) = 0.

(9)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

152

For any K > 0, there exists σK > 0 and a function FK in L2 (Q) such that |f (x, t, s)| ≤ FK (x, t) + σK |s|

(10)

for almost every (x, t) ∈ Q, for every s such that |s| ≤ K. u0 is a measurable function defined on Ω such that b(x, u0 ) ∈ L1 (Ω).(11) Remark 2.1. As already mentioned in the introduction, Problems (1)(3) does not admit a weak solution under assumptions (7)-(11) since the growths of b(x, u), Φ(u) and f (x, t, u) are not controlled with respect to u (so that these fields are not in general defined as distributions, even when u belongs L2 (0, T ; H01 (Ω))). Throughout this paper and for any non negative real number K we denote by TK (r) = min(K, max(r, −K)) the truncation function at height K. The definition of a renormalized solution for Problems (1)-(3) can be stated as follows. Definition 2.1. A measurable function u defined on Q is a renormalized solution of Problems (1)-(3) if TK (u) ∈ L2 (0, T ; H01 (Ω)) for any K ≥ 0 and b(x, u) ∈ L∞ (0, T ; L1(Ω)), (12) Z A(x, t)Du.Du dx dt −→ 0 as n → +∞, (13) {(t,x)∈Q ; n≤|u(x,t)|≤n+1}

and if, for every a function S in W 2,∞ (IR) increasing, which is piecewise C 1 and such that S 0 has a compact support, we have ∂bS (x, u) − div S 0 (u)A(x, t) + S 00 (u)A(x, t)Du.Du (14) ∂t − div S 0 (u)Φ(u) + S 00 (u)Φ(u)Du + f (x, t, u)S 0 (u) = 0 in D0 (Q),

where bS (x, r) =

Z

bS (x, u)(t = 0) = bS (x, u0 ) in Ω, r 0

(15)

∂b(x, s) 0 S (s) ds. ∂s

The following remarks are concerned with a few comments on definition 2.1. Remark 2.2. Equation (14) is formally obtained through pointwise multiplication of equation (1) by S 0 (u). Note that due to (12) each term in (14) has a meaning in L1 (Q) + L2 (0, T ; H −1(Ω)).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

153

Indeed, if K ≥ 0 is such that suppS 0 ⊂ [−K, K], the following identifications are made in (14) : ? bS (x, u) belongs to L∞ (Q) because |bS (x, u)| ≤ kAK kL∞ (Ω) kSkL∞(IR) . ? S 0 (u)A(x, t)Du identifies with S 0 (u)A(x, t)DTK (u) a.e. in Q. Since indeed |TK (u)| ≤ K a.e. in Q, assumptions (6) and (12) imply that S 0 (u)A(x, t)DTK (u) ∈ L2 (Q)N . ? S 00 (u)A(x, t)DuDu identifies with S 00 (u)A(x, t)DTK (u).DTK (u) and in view of (6), and (12) one has S 00 (u)A(x, t)DTK (u).DTK (u)) ∈ L1 (Q). ? S 0 (u)Φ(u) and S 00 (u)Φ(u)∇u respectively identify with S (u)Φ(TK (u)) and S 00 (u)Φ(TK (u))∇TK (u). Due to the properties of S and (7), the functions S 0 , S 00 and Φ ◦ TK are bounded on IR so that (12) implies that S 0 (u)Φ(TK (u)) ∈ (L∞ (Q))N , and S 00 (u)Φ(TK (u))DTK (u) ∈ L2 (Q). ? f (x, t, u)S 0 (u) identifies with f (x, t, TK (u))S 0 (u) and in view of (10) one has f (x, t, TK (u))S 0 (u) ∈ L2 (Q). The above considerations show that equation (14) takes place in D 0 (Q) (x,u) belongs to L1 (Q) + L2 (0, T ; H −1 (Ω)). Due to the propand that ∂bS∂t erties of S and (5), we have bS (x, u) belongs to L2 (0, T ; H01 (Ω)) (see (17)), which implies that bS (x, u) belongs to C 0 ([0, T ]; L1 (Ω)) (for a proof of this trace result see 20 ), so that the initial condition (15) makes sense. 0

Remark 2.3. Due to the properties of S (increasing) and (5), we have λK |S(r)−S(r0 )| ≤ bS (x, r)−bS (x, r0 ) ≤ AK (x)|S(r)−S(r0 )| ∀r, r0 ∈ IR (16) and |DbS (x, u)| ≤ kAK kL∞(Q) kSkL∞(IR) |DTK (u)| + K|BK (x)|kS 0 kL∞ (IR) . (17) 3. Existence result This section is devoted to establish the following existence theorem. Theorem 3.1. Under assumptions (7)-(11) there exists at least a renormalized solution u of Problems (1)-(3). Proof of Theorem 3.1. The proof is divided into 7 steps.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

154

? Step 1 : Approximate problem. Let us introduce the following regularization of the data: for ε > 0 fixed bε (x, s) = b(x, T ε1 (s)))

a.e. in Ω, ∀s ∈ IR.

Φε is a lipschitz-continuous bounded function from IR into IR N

(18) (19)

such that Φε uniformly converges to Φ on any compact subset of IR as ε tends to 0. f ε (x, t, s) = f (x, t, T ε1 (s))

a.e. in Q, ∀s ∈ IR.

uε0 ∈ C0∞ (Ω) : bε (x, uε0 ) −→ b(x, u0 ) in L1 (Ω) as ε tends to 0.

(20) (21)

Let us now consider the following regularized problem: ∂bε (x, uε ) − div A(x, t)Duε + Φε (uε )Duε + f ε (x, t, uε ) = 0 in Q,(22) ∂t uε = 0 on (0, T ) × ∂Ω,

(23)

bε (x, uε )(t = 0) = bε (x, uε0 ) in Ω.

(24)

In view of (18), bε satisfy (4) and (5), and due to (5), there exists λε > 0, a function Aε in L∞ (Ω) and a function Bε in L2 (Ω) such that ∂b (x, s) ∂bε (x, s) ε ≤ Aε (x) and ∇x λε ≤ ≤ Bε (x) a.e. in Ω, ∂s ∂s (25) ε ∀s ∈ IR. In view of (20), f satisfy (8), (9) and (10), and due to (10), there exists σε > 0 and a function Fε in L2 (Q) such that |f ε (x, t, s)| ≤ Fε (x, t) + σε |s|

(26)

As a consequence, proving existence of a weak solution uε ∈ L2 (0, T ; H01(Ω)) of (22)-(24) is an easy task (see e.g. 15 ). ? Step 2 : A priori estimates. The estimates derived in this step rely on usual techniques for problems of type (22)-(24) and we just sketch the proof of them (the reader is referred to ,2 ,3 ,7 ,4 5 or to ,8 ,18 19 for elliptic versions of (22)-(24)). Using TK (uε ) as a test function in (22) leads to Z tZ Z bεK (x, uε )(t) dx + A(x, t)Duε .DTK (uε ) dx ds (27) Ω

0

Ω

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

155

+ =

Z tZ Z0

Ω

ε

ε

Φε (u )DTK (u ) dx ds +

Z tZ 0

Ω

bεK (x, uε0 ) dx

f ε (x, t, uε )TK (uε ) dx ds Ω

Z

r

∂bε (x, s) ds. ∂s 0 The Lipschitz character of Φε , Stokes formula together with the boundary condition (24) make it possible to obtain Z tZ Φε (uε )DTK (uε ) dx ds = 0, (28)

for almost every t in (0, T ), and where bεK (x, r) =

0

for almost any t ∈ (0, T ).

TK (s)

Ω

Due to the definition of bεK we have 0 ≤ Z K |bε (x, uε0 )| dx.

Z

Ω

bεK (x, uε0 ) dx ≤

Ω

Observing that both terms on the left hand side of the above equality are nonnegative, and since A(x, t) satisfies (6), the properties of bε (x, uε0 ), permit to deduce from (27) that TK (uε ) is bounded in L2 (0, T ; H01(Ω))

(29)

independently of ε for any K ≥ 0. Proceeding as in ,3 4 and 7 that for any S ∈ W 2,∞ (IR) such that S 0 is compact (suppS 0 ⊂ [−K, K]) bS (x, uε ) is bounded in L2 (0, T ; H01 (Ω))

(30)

∂bS (x, uε ) is bounded in L1 (Q) + L2 (0, T ; H −1 (Ω)) ∂t

(31)

and

independently of ε. As a consequence of (17), (25) and (29) we then obtain (30). To show that (31) holds true, we multiply the equation for uε in (22) by S 0 (uε ) to obtain ∂bεS (x, uε ) (32) = div S 0 (uε )A(x, t)Duε ∂t −S 00 (uε )A(x, t)Duε .Duε + div(Φε (uε ))S 0 (uε ) − f ε (x, t, uε )S 0 (uε ) = 0 Z r ε ∂b (x, s) 0 S (s) ds. in D0 (Q), where bεS (x, r) = ∂s 0

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

156

Since suppS 0 and suppS 00 are both included in [−K, K], uε may be replaced by TK (uε ) in each of these terms. As a consequence, each term in the right hand side of (32) is bounded either in L2 (0, T ; H −1 (Ω)) or in L1 (Q). (see ,4 7 ). As a consequence of (6), (7), (10) and (29) we then obtain (31). For any integer n ≥ 1, consider the Lipschitz-continuous function θn defined through θn (r) = Tn+1 (r) − Tn (r). Remark that kθn kL∞ (IR) ≤ 1 for any n ≥ 1 and that θn (r) → 0 for any r when n tends to infinity. Using that admissible test function θn (uε ) in (22) leads to Z Z tZ ε bε,n (x, u )(t) dx + A(x, t)Duε .Dθn (uε ) dx ds (33) Ω

+ =

Z tZ Z0

Ω

0

Ω

Φε (uε )Dθn (uε ) dx ds = Ω

Z tZ 0

bε,n (x, uε0 ) dx,

f ε (x, uε )θn (uε ) dx ds Ω

for almost any t in (0, T ) and where bε,n (x, r) = The Lipschitz character of Φε , and since

Z

r 0

∂bε (x, s) θn (s) ds. ∂s

bε,n (x, r) ≥ 0, f ε (x, t, uε )θn (uε ), equality (33) implies that Z tZ Z ε ε A(x, t)Du .Dθn (u ) dx ds ≤ bε,n (x, uε0 ) dx, 0

Ω

(34)

Ω

for almost t ∈ (0, T ). ? Step 3 : Limit of the approximate solutions. Arguing again as in ,3 ,4 5 and 7 estimates (30) and (31) imply that, for a subsequence still indexed by ε, bε (x, uε ) converges strongly in L1 (Q) and almost every where to b(x, u) (35) in Q and with the help of (16) and (29), uε converges almost every where to u in Q,

(36)

TK (uε ) converges weakly to TK (u) in L2 (0, T ; H01 (Ω)),

(37)

θn (uε ) * θn (u) weakly in L2 (0, T ; H01 (Ω))

(38)

as ε tends to 0 for any K > 0 and any n ≥ 1.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

157

We now establish that b(x, u) belongs to L∞ (0, T ; L1 (Ω)). Indeed using as a test function in (22) and letting σ go to zero, it follows that Z |bε (x, uε )|(t) dx ≤ kbε (x, uε0 )kL1 (Ω) a.e. in (0, T ). (39)

1 ε σ Tσ (u )

Ω

With of (21) and (35), we have b(x, u) belongs to L∞ (0, T ; L1(Ω)). We are now in a position to exploit (34). Due to the definition of θn , the pointwise convergence of uε to u and bε (x, uε0 ) to b(x, u0 ) then imply that Z Z bn (x, u0 ) dx. lim A(x, t)Duε .Duε dx dt ≤ ε→0

{n≤|uε |≤n+1}

Ω

Since θn converge to zero everywhere as n goes to zero. The Lebesgue’s convergence theorem permits to conclude that Z A(x, t)DTn+1 (uε ).DTn+1 (uε ) dx dt = 0. (40) lim lim n→+∞ ε→0

{n≤|uε |≤n+1}

2 1 1 Since A(x, t)DTn+1 (uε ).DTn+1 (uε ) = A(x, t) 2 DTn+1 (uε ) (with A 2 de1

1

notes the coercive symmetric matrix such that A 2 .A 2 = A) and (37), (6) 1 1 imply that A(x, t) 2 DTn+1 (uε ) * A(x, t) 2 DTn+1 (u) weakly in (L2 (Q))N , and (13) is then established.

? Step 4 : Time regularization. This step is devoted to introduce for K ≥ 0 fixed, a time regularization of the function TK (u) in order to perform the monotonicity method which will be developed in Step 5 and Step 6. This kind of regularization has been first introduced by R. Landes (see Lemma 6 and Proposition 3, p. 230 and Proposition 4, p. 231 in 14 ). More recently, it has been exploited in 5 and 13 to solve a few nonlinear evolution problems with L1 or measure data. This specific time regularization of TK (u) (for fixed K ≥ 0) is defined as follows. Let (v0µ )µ be a sequence of functions defined on Ω such that v0µ ∈ L∞ (Ω) ∩ H01 (Ω) for all µ > 0, kv0µ kL∞ (Ω) ≤ K v0µ → TK (u0 ) a.e. in Ω and

∀µ > 0,

1 µ kv kL2 (Ω) → 0, as µ → +∞. µ 0

(41) (42) (43)

Existence of such a subsequence (v0µ )µ is easy to establish (see e.g. 21 ). For fixed K ≥ 0 and µ > 0, let us consider the unique solution TK (u)µ ∈

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

158

L∞ (Q) ∩ L2 (0, T ; H01 (Ω)) of the monotone problem: ∂TK (u)µ + µ TK (u)µ − TK (u) = 0 in D0 (Q). ∂t

(44)

TK (u)µ (t = 0) = v0µ in Ω.

(45)

Remark that due to (44), we have for µ > 0 and K ≥ 0, ∂TK (u)µ ∈ L2 (0, T ; H01 (Ω)). ∂t The behavior of TK (u)µ as µ → +∞ is investigated in ) and we just recall here that (44)-(45) imply that

(46) 14

(see also ,5

13

and

21

TK (u)µ → TK (u) a.e. in Q ; and in L∞ (Q) weak ? and strongly in L2 (0, T ; H01 (Ω)) as µ → +∞. kTK (u)µ kL∞(Q) ≤ max kTK (u)kL∞ (Q) ; kv0µ kL∞ (Ω) ≤ K

(47)

(48)

for any µ and any K ≥ 0. Let h ∈ W 1,∞ (IR), h ≥ 0, supp h is compact. The main estimate is Lemma 3.1. Z lim lim µ→+∞ ε→0

T 0

Z

s 0

E D ∂b (x, uε ) ε , h(uε ) TK (uε ) − (TK (u))µ dt ds ≥ 0 ∂t

where h , i denotes the duality pairing between L1 (Ω) + H −1 (Ω) and L∞ (Ω) ∩ H01 (Ω). Proof of Lemma 3.1 : The Lemma is proved in .22 ? Step 5. In this step we prove the following lemma which is the key point in the monotonocity arguments that will be developed in Step 6. Lemma 3.2. The subsequence of uε defined is Step 3 satisfies for any K ≥0 Z T Z tZ lim A(x, t)DTK (uε ).DTK (uε ) dx ds dt (49) ε→0

≤

0

Z

0

T 0

Z tZ 0

Ω

A(x, t)DTK (u).DTK (u) dx ds dt. Ω

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

159

Proof of Lemma 3.2: We first introduce a sequence of increasing C ∞ (IR)functions Sn such that, for any n ≥ 1, Sn (r) = r for |r| ≤ n, supp(Sn0 ) ⊂ [−(n + 1), (n + 1)] and kSn00 kL∞ (IR) ≤ 1. We use the sequence TK (u)µ of approximations of TK (u) defined by (44), (45) of Step 4, and plug the test function Sn0 (uε )(TK (uε ) − TK (u)µ ) (for ε > 0 and µ > 0) in (22). Through setting, for fixed K ≥ 0, Wµε = (TK (uε ) − TK (u)µ ) we obtain upon integration over (0, t) and then over (0, T ): Z T Z tD E ∂bε (x, uε ) , Sn0 (uε )Wµε ds dt ∂t 0 0 +

+

Z

Z

Z tZ

T 0

0

Z tZ

T

0

0

+

Z

T

0

Z

+

+

Z

T 0

T 0

Ω

Ω

Sn00 (uε )Wµε A(x, t)Duε .Duε dx ds dt

0

Z tZ

Z tZ 0

Ω

(51)

Sn0 (uε )A(x, t)Duε .DWµε dx ds dt

Z tZ

0

(50)

Ω

Ω

Φε (uε )Sn0 (uε )DWµε dx ds dt

Sn00 (uε )Wµε Φε (uε )Duε dx ds dt

f ε (x, t, uε )Sn0 (uε )Wµε dx ds dt = 0.

In the following we pass to the limit in (51) as ε tends to 0, then µ tends to +∞ and then n tends to +∞, the real number K ≥ 0 being kept fixed. In order to perform this task we prove below the following results for fixed K ≥ 0: Z T Z tD E ∂bε (x, uε ) lim lim , Sn0 (uε )Wµε ds dt ≥ 0 for any n ≥ K, µ→+∞ ε→0 0 ∂t 0 (52) Z T Z tZ Sn0 (uε )Φε (uε )DWµε dx ds dt = 0 for any n ≥ 1, (53) lim lim µ→+∞ ε→0

lim lim

µ→+∞ ε→0

0

Z

0

T 0

Ω

Z tZ 0

Ω

Sn00 (uε )Wµε Φε (uε )Duε dx ds dt = 0 for any n, (54)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

160

lim

Z lim

lim

n→+∞ µ→+∞ ε→0

and lim lim

µ→+∞ ε→0

Z

T 0

Z tZ

T 0

Z tZ 0

0

Ω

Ω

Sn00 (uε )Wµε A(x, t)Duε .Duε dx ds dt = 0,

(55)

f ε Sn0 (uε )Wµε dx ds dt = 0 for any n ≥ 1.

(56)

Proof of (52). In view of the definition (50) of Wµε , lemma 3.1 applies with h = Sn for fixed n ≥ K. As a consequence (52) holds true. Proof of (53). For fixed n ≥ 1, we have

Sn0 (uε )Φε (uε )DWµε = Sn0 (uε )Φε (Tn+1 (uε ))DWµε suppSn0

(57)

Sn0

a.e. in Q, and where ⊂ [−(n + 1), n + 1]. Since is smooth 0 ε ε and bounded, (19) and (36) lead to Sn (u )Φε (Tn+1 (u )) converges to Sn0 (u)Φ(Tn+1 (u)) a.e. in Q and in L∞ (Q) weak ?, as ε tends to 0. For fixed µ > 0, we have Wµε * (TK (u) − TK (u)µ ) weakly in L2 (0, T ; H01 (Ω))

(58)

∞

and a.e. in Q and in L (Q) weak ?, as ε tends to 0. As a consequence of (57) and (58) we deduce that Z T Z tZ lim Sn0 (uε )Φε (uε )DWµε dx ds dt (59) ε→0

=

Z

T 0

0

Z tZ 0

Ω

0

Ω

i h Sn0 (u)Φ(u)D TK (u) − TK (u)µ dx ds dt

for any µ > 0. Appealing now to (47) and passing to the limit as µ → +∞ in (59) allows to conclude that (53) holds true. Proof of (54). For fixed n ≥ 1, and by the same arguments that those that lead to (53), we have Sn00 (uε )Φε (uε )Duε Wµε = Sn00 (uε )Φε (Tn+1 (uε ))DTn+1 (uε )Wµε a.e. in Q. From (19) and (36), it follows that for any µ > 0 Z T Z tZ lim Sn00 (uε )Φε (uε )Duε Wµε dx ds dt ε→0

=

Z

T 0

0

Z tZ 0

Ω

0

Ω

Sn00 (uε )Φε (Tn+1 (uε ))DTn+1 (uε )Wµε dx ds dt

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

161

with the help of (58) passing to the limit, as µ tends to +∞, in the above equality leads to (53). Proof of (55). For any n ≥ 1 fixed, we have suppSn00 ⊂ [−(n + 1), −n] ∪ [n, n + 1]. As a consequence Z T Z tZ Sn00 (uε )A(x, t)Duε .Duε Wµε dx ds dt 0

0

Ω

≤ TC

Z

A(x, t)Duε .Duε dx dt,

{n≤|uε |≤n+1}

for any n ≥ 1, and any µ > 0, where C is a constant independent of n, µ. With the help of (40) passing to the limit, as ε tends to zero, µ tends to +∞ and n tends to +∞ and to establish (55). Proof of (56). For fixed n ≥ 1, and in view (20), (36) and (58), Lebesgue’s convergence theorem implies that for any µ > 0 and any n ≥ 1 Z T Z tZ lim f ε (x, t, Tn+1 (uε ))Sn0 (uε )Wµε dx ds dt ε→0

=

Z

T 0

0

Z tZ 0

Ω

0

Ω

f (x, t, Tn+1 (u))Sn0 (u) TK (u) − TK (u)µ dx ds dt.

Now for fixed n ≥ 1, using (47) permits to pass to the limit as µ tends to +∞ in the above equality to obtain (56). We now turn back to the proof of lemma 3.2, due to (51), (52), (53), (54), (55) and (56), we are in a position to pass to the lim-sup when ε tends to zero, then to the limit-sup when µ tends to +∞ and then to the limit as n tends to +∞ in (51). We obtain using the definition of Wµε that for any K≥0 Z T Z tZ Sn0 (uε )A(x, t) lim lim lim n→+∞ µ→+∞ ε→0 0 0 Ω ×Duε D TK (uε ) − TK (u)µ dx ds dt ≤ 0. Since Sn0 (uε )DTK (uε ) = DTK (uε ) for K ≤ n. The above inequality implies that for K ≤ n Z T Z tZ lim A(x, t)Duε .DTK (uε ) dx ds dt ε→0

≤ lim

lim

lim

n→+∞ µ→+∞ ε→0

0

Z

0

T 0

Z tZ 0

(60)

Ω

Ω

Sn0 (uε )A(x, t)DTn+1 (uε ).DTK (u)µ dx ds dt.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

162

The right hand side of (61) is computed as follows. Due to (37) it follows that for fixed n ≥ 1 Sn0 (uε )DTn+1 (uε ) * Sn0 (u)DTn+1 (u) weakly in (L2 (Q))N

when ε tends to 0. The strong convergence of TK (u)µ to TK (u) in L2 (0, T ; H01 (Ω)) as µ tends to +∞, then allows to conclude that Z T Z tZ lim lim Sn0 (uε )A(x, t)DTn+1 (uε ).DTK (u)µ dx ds dt (61) µ→+∞ ε→0

=

0

Z

0

Ω

Z tZ

T 0

0

Z

=

T 0

Ω

Sn0 (u)A(x, t)DTn+1 (u).DTK (u) dx ds dt

Z tZ 0

A(x, t)DTK (u).DTK (u) dx ds dt

Ω Sn0 (r)

as soon as K ≤ n, since = 1 for |r| ≤ n. Recalling (60) and (61) allows to conclude (49) holds true and the proof of lemma 3.2 is complete. ? Step 6 : The strong convergence of truncates. In this step we prove the following monotonicity estimate : Lemma 3.3. The subsequence of uε defined in step 3 satisfies for any K ≥0 Z T Z tZ lim A(x, t)[DTK (uε )−DTK (u)].[DTK (uε )−DTK (u)]dx dt ds = 0. ε→0

0

0

Ω

(62)

And TK (uε ) −→ TK (u) strongly in L2 (0, T, H01 (Ω))

(63)

as ε goes to zero. Proof of Lemma 3.3. Let K ≥ 0 be fixed. we have Z T Z tZ h ih i A(x, t) DTK (uε ) − DTK (u) . DTK (uε ) − DTK (u) dx dt ds 0

0

Ω

(64)

=

Z

−

T 0

Z

0

Z tZ 0

T

A(x, t)DTK (uε ).DTK (uε ) dx dt ds Ω

Z tZ 0

A(x, t)DTK (u)DTK (uε ) dx dt ds Ω

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

163

−

Z

T

0

Z tZ 0

Ω

h i A(x, t) DTK (uε ) − DTK (u) .DTK (u) dx dt ds.

Using (49) of lemma 3.2, we obtain Z T Z tZ lim ≤ A(x, t)DTK (uε ).DTK (uε ) dx dt ds ε→0

0

≤

Z

T 0

0

Z tZ 0

A(x, t)DTK (u).DTK (u) dx dt ds. Ω

As a consequence of (37) we have for all K > 0 Z T Z tZ h i lim A(x, t) DTK (uε ) − DTK (u) .DTK (u) dx dt ds = 0 ε→0

0

0

(65)

Ω

(66)

Ω

(65) and (66) allow to pass to the lim-sup as ε tends to zero in (64) and to obtain (62) and (63) of lemma 3.3. ? Step 7. In this step, u is shown to satisfies (14) and (15). Let S be a function in W 2,∞ (IR) such that S 0 has a compact support. Let K be a positive real number such that suppS 0 ⊂ [−K, K]. Pointwise multiplication of the approximate equation (22) by S 0 (uε ) leads to ∂bεS (x, uε ) − div S 0 (uε )A(x, t)Duε + S 00 (uε )A(x, t)Duε .Duε (67) ∂t

− div S 0 (uε )Φε (uε ) +S 00 (uε )Φε (uε )Duε +f ε (x, t, uε )S 0 (uε ) = 0 in D0 (Q). Z

r

∂bε (x, s) 0 S (s) ds. In what follows we pass to the limit ∂s 0 as ε tends to 0 in each term of (67). where

bεS (x, r)

=

? Since S is bounded, and bεS (x, uε ) converges to bS (x, u) a.e. in Q and in ∂bε (x,uε ) (x,u) converges to ∂bS∂t in D0 (Q) as ε tends to L∞ (Q) weak ?. Then S ∂t 0. ? Since suppS 0 ⊂ [−K, K], we have forS 0 (uε )A(x, t)Duε = S 0 (uε )A(x, t)DTK (uε ) a.e. in Q. The pointwise convergence of uε to u as ε tends to 0, the bounded character of S 0 and (37) imply that S 0 (uε )A(x, t)DTK (uε ) * S 0 (u)A(x, t)DTK (u) weakly in (L2 (Q))N , as ε tends to 0. And the term S 0 (u)A(x, t)DTK (u) = S 0 (u)A(x, t)Du a.e. in Q.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

164

? Since suppS 00 ⊂ [−K, K], we have S 00 (uε )A(x, t)Duε .Duε = A(x, t)DS 0 (uε ).DTK (uε ) a.e. in Q. Due to (29) and (36), DS 0 (uε ) converges to DS 0 (u) weakly in (L2 (Q))N as ε tends to 0, and (63) of lemma 3.3 allow to conclude that A(x, t)DS 0 (uε ).DTK (uε ) * A(x, t)DS 0 (u).DTK (u) weakly in L1 (Q), as ε tends to 0. And A(x, t)DS 0 (u).DTK (u) = S 00 (u)A(x, t)Du.Du a.e. in Q. ? Since suppS 0 ⊂ [−K, K], we have S 0 (uε )Φε (uε ) = S 0 (uε )Φε (TK (uε )) a.e. in Q. As a consequence of (19) and (36), it follows that for any 1 ≤ q < +∞, S 0 (uε )Φε (uε ) converges to S 0 (u)Φ(TK (u)) strongly in Lq (Q), as ε tends to 0. The term S 0 (u)Φ(TK (u)) is denoted by S 0 (u)Φ(u). ? Since S 0 ∈ W 1,∞ (IR) with suppS 0 ⊂ [−K, K], we have S 00 (uε )Φε (uε )Duε = Φε (TK (uε ))DS 0 (uε ) a.e. in Q, we have, DS 0 (uε ) conN verges to DS 0 (u) weakly in L2 (Q) as ε tends to 0, while Φε (TK (uε )) is uniformly bounded with respect to ε and converges a.e. in Q to Φ(TK (u)) as ε tends to 0. Therefore S 00 (uε )Φε (uε )Duε * Φ(TK (u))DS 0 (u) weakly in L2 (Q). ? Due to (20) and (35), we have f ε (x, t, Tn+1 (uε ))S 0 (uε ) converges to f (x, t, Tn+1 (u))S 0 (u) strongly in L1 (Q), as ε tends to 0. As a consequence of the above convergence result, we are in a position to pass to the limit as ε tends to 0 in equation (67) and to conclude that u satisfies (28). It remains to show that bS (x, u) satisfies the initial condition (29). To this end, firstly remark that, S ∈ W 2,∞ (IR) such that S 0 has a compact support, as a consequence of (17) we have bS (x, uε ) is bounded in L2 (0, T ; H01 (Ω)). Secondly, (67) and the above considerations on the be∂bεS (x, uε ) is bounded havior of the terms of this equation show that ∂t 1 2 −1 in L (Q) + L (0, T ; H (Ω)). As a consequence, an Aubin’s type lemma (see, e.g, ,24 Corollary 4) implies that bS (x, uε) lies in a compact set of C 0 ([0, T ]; W −1,s (Ω)) for any s < inf 2, NN−1 . It follows that, on one

hand, bS (x, uε )(t = 0) = bS (x, uε0 ) converges to bS (x, u)(t = 0) strongly in W −1,s (Ω). On the other hand, (21) and the smoothness of S imply that bS (x, uε0 ) converges to bS (x, u)(t = 0) strongly in Lq (Ω) for all q < +∞.Then we conclude that bS (x, u)(t = 0) = bS (x, u0 ) in Ω. As a conclusion of step 1-step 7, the proof of theorem 3.1 is complete.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

165

4. Comparison principle and uniqueness result This section is concerned with a comparison principle (and a uniqueness result) for renormalized solutions in the case where f (x, t, u) is independent of u. We establish the following theorem. Theorem 4.1. Assume that assumptions (4), (5), (6), (7) and (11) hold true and moreover that For any K > 0, there exists a positive real number βK > 0, such that ∂b(x, z ) ∂b(x, z ) 1 2 − (68) ≤ βK z1 − z2 ∂s ∂s for almost every x in Ω, and for every z1 and every z2 such that |z1 | ≤ K and |z2 | ≤ K. Φ is a locally lipschitz-continuous function on IR.

(69)

Let then u1 and u2 be renormalized solutions corresponding to the data (f1 , u10 ) and (f2 , u20 ) for problem (i = 1, 2) ∂b(x, ui ) − div A(x, t)Dui + Φ(ui ) = fi (x, t) in Ω × (0, T ), (70) ∂t b(x, ui )(t = 0) = b(x, ui0 ) in Ω,

(71)

ui = 0 on ∂Ω × (0, T ),

(72)

f1 , f2 ∈ L1 (Ω × (0, T )).

(73)

If these data satisfying f1 ≤ f2 and u10 ≤ u20 almost every where, we have u1 ≤ u2 almost every where. Sketch of the Proof of theorem 4 .1 . Here we give just an idea on how u1 ≤ u2 can be obtained following the outlines of .23 The proof is divided into two Steps. In Step 1, we define a smooth approximation Sn of Tn , and we consider tow renormalized solutions u1 and 2 u2 of (70)-(72) for the data (f1 , u10 ) and (f 2 , u0 ) respectively we plug the test function σ1 Tσ + bSn (x, u1 ) − bSn (x, u2 )) in the difference of equations (14) for u1 and u2 in which we have taken S = Sn . In Step 2, we investigate the behavior of the different terms in the estimate obtained in step 1 (estimates (76)) as σ tends to 0 and when n tends to +∞.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

166

? Step 1 . Remark that when Φ is locally-continuous on IR the following derivation is licit for any function S and u satisfying the conditions mentioned in Definition 2.1. div S 0 (u)Φ(u) − S 00 (u)Φ(u)Du = S 0 (u)Φ0 (u)Du = div(ΦS (u)) (74) Where ΦS = (ΦS,1 , ΦS,2 , · · · , ΦS,N ) with Z r ΦS,i (r) = Φ0S,i (t)S 0 (t) dt. 0

Let us now introduce a specific choice of function S in (14). For all n > 0, let Sn ∈ C 1 (IR) be the function defined by Sn0 (r) = 1 for |r| ≤ n ; Sn0 (r) = n + 1 − |r| for n ≤ |r| ≤ n + 1 and Sn0 (r) = 0 for |r| ≥ n + 1. It yields, taking S = Sn in (14) ∂bSn (x, ui ) (75) − div S 0 (ui )A(x, t)Dui + S 00 (ui )A(x, t)Dui Dui ∂t − div ΦSn (ui ) = fi Sn0 (ui ) in D0 (Q) ; Z r ∂b(x, s) 0 Sn (s) ds. for i = 1, 2 and where bSn (x, r) = ∂s 0 1 We use Tσ+ bSn (x, u1 )−bSn (x, u2 ) as a test function in the difference σ of equations (75) for u1 and u2 . Z T Z t ∂ bS (x, u1 ) − bS (x, u2 ) n n 1 ; Tσ+ bSn (x, u1 ) − bSn (x, u2 ) i ds dt h σ 0 0 ∂t (76) +Aσn = Bnσ + Cnσ + Dnσ for any σ > 0, n > 0, and where Z Z Z h i 1 T t σ Sn0 (u1 )A(t, x)Du1 − Sn0 (u2 )A(t, x)Du2 An = σ 0 0 Ω

Bnσ =

−

1 σ

Z

1 σ T 0

Z

T 0

Z

Z tZ 0

0

tZ

Ω

DTσ+ bSn (x, u1 ) − bSn (x, u2 ) dx ds dt

Ω

(77)

Sn00 (u1 )A(x, t)Du1 Du1 Tσ+ bSn (x, u1 ) − bSn (x, u2 ) dx ds dt (78)

Sn00 (u2 )A(x, t)Du2 Du2 Tσ+ bSn (x, u1 ) − bSn (x, u2 ) dx ds dt

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

167

1 σ

Z

1 σ Dn = σ

Z

Cnσ =

T

0

0

T

Z

0

Z

tZ

0

Ω

tZ

Ω

h

i

ΦSn (u1 ) − ΦSn (u2 ) DTσ+ bSn (x, u1 ) − bSn (x, u2 ) dx ds dt

h

i

(79)

f1 Sn0 (u1 ) − f2 Sn0 (u2 ) Tσ+ bSn (x, u1 ) − bSn (x, u2 ) dx ds dt. (80)

In the sequel we pass to the limit in (76) when σ tends to 0 and then n tends to +∞. Upon application of lemma 2.4 of ,9 the first term in the right hand side of (76) is derived as Z T Z t ∂ bS (x, u1 ) − bS (x, u2 ) n n 1 ; Tσ+ bSn (x, u1 ) − bSn (x, u2 ) i ds dt h σ 0 0 ∂t (81) R = σ1 Q T˜σ+ bSn (x, u1 ) − bSn (x, u2 ) dx dt R − Tσ Ω T˜σ+ bSn (x, u10 ) − bSn (x, u20 ) dx where T˜σ+ (t) =

Z

t

0

Tσ+ (s) ds.

Due to the assumption u10 ≤ u20 a.e. in Ω and the monotone character of bSn (x, .) and Tσ (.) , we have Z (82) T˜σ+ bSn (x, u10 ) − bSn (x, u20 ) dx = 0 Ω

It follows from (76), (81) and (82) that Z 1 T˜σ+ bSn (x, u1 ) − bSn (x, u2 ) dx dt + Aσn = Bnσ + Cnσ + Dnσ σ Q

(83)

for any σ > 0 and any n > 0.

? Step 2 . In this step, we study the behaviors of the terms Aσn , Bnσ , Cnσ and Dnσ when σ tends to 0 and n → +∞. More precisely, we prove the following Lemma Lemma 4.1. We have lim lim Aσn ≥ 0,

(84)

lim Bnσ = 0,

(85)

n→+∞ σ→0

lim

n→+∞ σ→0

lim Cnσ = 0

σ→0

for all n,

(86)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

168

lim

lim Dnσ ≤ 0.

n→+∞ σ→0

(87)

Proof of Lemma 4.1. The lemma is proved in .23 In view of estimates (82), (83), (84), (85), (86) and (87) we have Z + b(x, u1 ) − b(x, u2 ) dx dt ≤ 0, Q

so that b(x, u1 ) ≤ b(x, u2 ) a.e. in Q which in turn implies that u1 ≤ u2 a.e. in Q, theorem 4.1 will be then established. References 1. P. B´ enilan, L. Boccardo, T. Gallou¨ et, R. Gariepy, M. Pierre and J.-L. Vazquez, An L1 -theory of existence and uniqueness of solutions of nonlinear elliptic equations, Ann. Scuola Norm. Sup. Pisa, 22, (1995), 241-273. 2. D. Blanchard, Truncation and monotonicity methods for parabolic equations equations, Nonlinear Anal., 21, (1993), 725-743. 3. D. Blanchard, and F. Murat, Renormalized solutions of nonlinear parabolic problems with L1 data, Existence and uniqueness, Proc. Roy. Soc. Edinburgh Sect., A 127, (1997), 1137-1152. 4. D. Blanchard, F. Murat and H. Redwane, Existence et unicit´e de la solution reormalis´ee d’un probl`eme parabolique assez g´en´eral, C. R. Acad. Sci. Paris S´er., I329, (1999), 575-580. 5. D. Blanchard, F. Murat and H. Redwane, Existence and Uniqueness of a Renormalized Solution for a Fairly General Class of Nonlinear Parabolic Problems, J. Differential Equations, 177, (2001), 331-374. 6. D. Blanchard and A. Porretta, A Stefan problems with nonlinear diffusion and convection, J. Diff. Equations, 210, (2005), 383-428. 7. D. Blanchard and H. Redwane, Renormalized solutions of nonlinear parabolic evolution problems, J. Math. Pure Appl., 77, (1998), 117-151. 8. L. Boccardo, D. Giachetti, J.-I. Diaz and F. Murat, Existence and regularity of renormalized solutions for some elliptic problems involving derivation of nonlinear terms, J. Differential Equations, 106, (1993), 215-237. 9. L. Boccardo, F. Murat and J.-P. Puel, Existence of bounded solutions for nonlinear elliptic unilateral problems, Ann. Mat. Pura Appl., 152, (1988), 183-196. 10. J. Carrillo, Entropy solutions for nonlinear degenerate problems, Arch. Ration. Mech. Anal., 147(4), (1999), 269-361. 11. J. Carrillo and P. Wittbold, Uniqueness of renormalized solutions of degenerate elliptic-parabolic problems, J. Differential Equations, 156, (1999), 93-121. 12. J. Carrillo and P. Wittbold, Renormalized entropy solution of a scalar conservation law with boundary condition, J. Differential Equations, 185(1), (2002), 137-160.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

169

13. A. Dall’Aglio and L. Orsina, Nonlinear parabolic equations with natural growth conditions and L1 data, Nonlinear Anal., 27, (1996), 59-73. 14. R. Landes, On the existence of weak solutions for quasilinear parabolic initial-boundary value problems, Proc. Roy. Soc. Edinburgh Sect., A89, (1981), 217-237. 15. J.-L. Lions, Quelques m´ethodes de r´esolution des probl`emes aux limites non lin´eaire, Dunod et Gauthier-Villars, Paris, (1969). 16. R.-J.DiPerna and P.-L. Lions, On the Cauchy problem for Boltzmann equations : Global existence and weak stability, Ann. Math., 130, (1989), 321-366. 17. J.-P. Lions, Mathematical Topics in Fluid Mechanics, Vol. 1 : Incompressible models, Oxford Univ. Press, (1996). 18. F. Murat, Soluciones renormalizadas de EDP elipticas non lineales, Cours a ` l’Universit´e de S´eville, Publication R93023, Laboratoire d’Analyse Num´erique, Paris VI, (1993). 19. F. Murat, Equations elliptiques non lin´eaires avec second membre L 1 S ou mesure, Comptes Rendus du 26`eme Congr`es National d’Analyse Num´erique Les Karellis, (1994), A12-A24, 20. A. Porretta, Existence results for nonlinear parabolic equations via strong convergence of trauncations, Ann. Mat. Pura ed Applicata, 177, (1999), 143172. 21. N. Grenon, R´esultats d’existence et comportement asymptotique pour des ´equations paraboliques quasi-lin´eaire, (1990). Th`ese Universit´e d’Orl´eans, France. 22. H. Redwane, Existence of a solution for a class of parabolic equations with three unbounded nonlinearities, Adv. Dyn. Syst. Appl., 2, (2007), pp. 241-264. 23. H. Redwane, Uniqueness of renormalized solutions for a class of parabolic equations with unbounded nonlinearities, Rendiconti di Matematica, VII, (2008), pp. 189-200. 24. J. Simon, Compact sets in Lp (0, T ; B), Ann. Mat. Pura Appl., 146, (1987), 65-96.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

170

Existence and uniqueness of solutions of some nonlinear equations in Orlicz spaces and weighted Sobolev spaces L. Aharouch Facult´ e polydisciplinaire Ouarzazate BP 638 Ouarzazate, Morocco E-mail: [email protected] A. Benkirane∗ , J. Bennouna† , and A. Touzani‡ D´ epartement de Math´ ematiques et Informatique Facult´ e des Sciences Dhar-Mahraz B.P 1796 Atlas F` es, Morocco E-mails: ∗ [email protected], † [email protected], ‡ [email protected] In this paper, we show, in the framework of Orlicz Sobolev spaces, the existence of renormalized and entropy solutions of some nonlinear equations and the existence and uniqueness of a unilateral problem. The existence of weak and renormalized solution of a nonlinear equation is also presented in the framework of weighted Sobolev spaces. Keywords: Orlicz Sobolev spaces; Weighted Sobolev spaces; Renormalized solution; Entropy solution; Weak solution; Boundary value problems.

1. Definition and existence of renormalized solutions in Orlicz space Let Ω be a bounded open subset of IRN with the segment property. Let M be an N-function satisfying the ∆2 -condition and let P be an N-function such that P << M . Let A : D(A) ⊂ W01 LM (Ω) → W −1 LM (Ω) be a mapping (not defined everywhere) given by A(u) = −div a(x, u, ∇u), where a : Ω × IR × IR N → IRN is a caratheodory function satisfying for a.e. x ∈ Ω and all t ∈ IR, ξ, ξ with ξ 6= ξ (1 − 1) (1 − 2)

| a(x, t, ξ) |≤ d(x) + k1 P

−1

M (k2 | t |) + k3 M

[a(x, t, ξ) − a(x, t, ξ)] [ξ − ξ] > 0,

−1

M (k4 | ξ |),

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

171

(1 − 3)

a(x, t, ξ) ξ ≥ αM (

|ξ| ), λ

where d(x) ∈ EM (Ω), d ≥ 0, α, λ ∈ IR∗+ , k1 , k2 , k3 , k4 ∈ IR+ . Consider the nonlinear elliptic problem (1 − 4)

−div a(x, u, ∇u) − divφ(u) + g(x, u) = f in Ω,

(1 − 5)

u = 0 on ∂Ω,

where f ∈ W −1 EM (Ω),

(1 − 6) and φ = (φ1 , . . . , φN ) satisfy

φ ∈ (C 0 (IR))N .

(1 − 7)

Let g(x, t) be a caratheodory function such that for a.e.x ∈ Ω and all t ∈ IR (1 − 8)

g(x, t) t ≥ 0, sup | g(., t) |= hn (.) ∈ L1 (Ω) ∀n.

(1 − 9)

|t|≤n

Note that no growth hypothesis is assumed on the function φ, which implies that for a solution u ∈ W01 LM (Ω) the term divφ(u) may be meaningless, even as a distribution. As in6 we define the following notion of renormalized solution, which gives a meaning to a possible solution of (1-4)-(1-5). The notion of renormalized solutions in the usual sens was introduced by R.J. Diperna and P.L. Lions10 for the study of the Boltzmann equations. This notion was then adapted to the study of the problem (1-1)-(1-2) by L. Boccardo, D. Giachetti, J.I. Diaz and F. Murat6 when the right hand 0 side is in W −1,p (Ω), by J.M. Rakotoson12 when the right hand side is in L1 (Ω), and finally by G. Dal Maso, F. Murat, L. Orsina and A. Prignet9 for the case of right hand side is general measure data. Definition 1.1. Assume that (1 − 1) − (1 − 3), (1 − 6) − (1 − 9) hold true. A function u is a renormalized solution of the problem (1 − 4) − (1 − 5) if u ∈ W01 LM (Ω), g(x, u) ∈ L1 (Ω), u g(x, u) ∈ L1 (Ω)

(1 − 10) (1 − 11)

(

0

−div a(x, u, ∇u) h(u) − div(φ(u)h(u)) + φ(u)h (u)∇u 0 +g(x, u)h(u) = f h(u) in D (Ω), ∀h ∈ Cc1 (IR).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

172

Remark 1.1. Let us note that in (1 − 11) every term is meaningful in the distributional sense (in contrast with (1 − 4)). Lemma 1.1. Let Ω be a bounded open subset of IR N with the segment property. If u ∈ (W01 LM (Ω))N then Z div u dx = 0. Ω

Theorem 1.1. Let M be an N-function satisfying the ∆2 -condition. Under assumptions (1 − 1) − (1 − 3), (1 − 6) − (1 − 9), there exists a renormalized solution u (in the sense of Definition 1.1) of problem (1 − 4) − (1 − 5). Proof of Theorem 1.1 see2 In the last theorem, we have supposed the N-function M satisfying the ∆2 -condition. In the next theorem we prove the same result without any restriction on the N-function M (i.e. without the ∆2 -condition). Theorem 1.2. Under assumptions (1 − 1) − (1 − 3), (1 − 6) − (1 − 9), there exists a renormalized solution u (in the sense of Definition 1.1) of problem (1 − 4) − (1 − 5). Proof of Theorem 1.2 see1

2. Definition and existence of entropy solutions in Orlicz space Let Ω be a bounded open subset of IRN with the segment property. Let M, P be two N-functions such that P << M . Let A : D(A) ⊂ W01 LM (Ω) → W −1 LM (Ω) be a mapping (not defined everywhere) given by: A(u) = −diva(x, u, ∇u) where a : Ω × IR × IR N → IRN is a caratheodory function satisfying for a.e. x ∈ Ω and all ξ, ξ ∈ IR N with ξ 6= ξ: (2 − 1) (2 − 2) (2 − 3)

| a(x, t, ξ) |≤ d(x) + k1 P

−1

M (k2 | t |) + k3 M

[a(x, t, ξ) − a(x, t, ξ)] [ξ − ξ] > 0 a(x, t, ξ) ξ ≥ αM (

|ξ| ) λ

−1

M (k4 | ξ |)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

173

where d(x) ∈ EM (Ω), d ≥ 0, α, λ ∈ IR∗+ , k1 , k2 , k3 , k4 ∈ IR+ . Consider the nonlinear elliptic problem (2 − 4)

−diva(x, u, ∇u) = f − divφ(u)

(2 − 5)

u = 0 on ∂Ω

where f ∈ L1 (Ω)

(2 − 6) and φ = (φ1 , . . . , φN ) satisfy (2 − 7)

φ ∈ (C 0 (IR))N .

Note that no growth hypothesis is assumed on the function φ, which implies that the term divφ(u) may be meaningless, even as a distribution. The notion of entropy solution, used in,7 allows us to give a meaning to a possible solution of (2-4)-(2-5). We introduce the following notation, see,1311 Definition 2.1. Let M be an N-function, we define the following set: 00

AM = {Q :

00

Q is an N-function such that QQ0 ≤ M M0 R. 1 −1 and 0 QoH ( 1− 1 )dr < ∞ where H(r) = r

N

M (r) r }

Remark 2.1. Let M (t) = tp and Q(t) = tq , then the condition Q ∈ AM , is equivalent to the following conditions: i) 2 − N1 < p < N , ii) q < p˜ = (p−1)N N −1 Definition 2.2. Assume that (2 − 1) − (2 − 3), (2 − 6) − (2 − 7) hold true, and suppose that AM 6= ∅. A function u is an entropy solution of problem (2 − 4) − (2 − 5) if  u ∈ W01 LQ (Ω) ∀Q ∈ AM ,    1 T (u) ∈ W L (Ω) ∀k > R0, 0 M Rk R  a(x, u, ∇u) ∇T [u − v]dx ≤ Ω f Tk [u − v]dx + Ω φ(u) ∇Tk [u − v]dx k  Ω  ∀v ∈ W01 LM (Ω) ∩ L∞ (Ω). Theorem 2.1. Assume that (2 − 1) − (2 − 3), (2 − 6) − (2 − 7) hold true, and suppose that AM 6= ∅, there exists an entropy solution u of problem (2 − 4) − (2 − 5) (in the sense of Definition 2.1).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

174

Proof of Theorem 2.1 see3 Remark 2.2. In the case M (t) = tp , our theorem gives a refinement of the regularity result (i.e. u ∈ W01,q (Ω), q < p˜ = (p−1)N N −1 ). 1 In fact by Theorem 2.1 we have u ∈ W0 LQ (Ω) ∀Q ∈ AM (for example p ˜ for Q(t) = logαt(e+t) , α > 1). 3. Existence and uniqueness of solution of unilateral problems with L1 -data in Orlicz spaces Let Ω be a bounded open subset of IRN with the segment property. Let M be an N-function, satisfying the ∆2 -condition. Let A : D(A) ⊂ W01 LM (Ω) → W −1 LM (Ω) be a mapping (not defined everywhere) given by: A(u) = −div a(x, ∇u) where a : Ω × IR N → IRN is a caratheodory function satisfying for a.e. x ∈ Ω and all ξ, ξ ∈ IR N with ξ 6= ξ: (3 − 1)

| a(x, ξ) |≤ d(x) + k1 M

−1

M (k2 | ξ |)

(3 − 2)

[a(x, ξ) − a(x, ξ)] [ξ − ξ] > 0

(3 − 3)

a(x, ξ) ξ ≥ αM (| ξ |)

where d(x) ∈ EM (Ω), d ≥ 0, α, k1 , k2 ∈ IR+ . Let Kψ = {v ∈ W01 LM (Ω) ∩ L∞ (Ω) : v ≥ ψ a.e. in Ω} where ψ : Ω → IR+ is a measurable function on Ω such that: ψ ∈ W01 LM (Ω) ∩ L∞ (Ω).

Assume that (3 − 4)

f ∈ L1 (Ω).

We suppose the regularity assumption on the obstacle function ψ: (3 − 5)

there exists ψ ∈ K such that ψ − ψ is continuous on Ω.

Consider the nonlinear elliptic unilateral problem:  u ∈ W01 LQ (Ω) ∀Q ∈ AM    u ≥ ψ a.e. in Ω (3 − 6)  T (u) ∈ W01 LM (Ω) ∀k R> 0   Rk a(x, ∇u) ∇T [u − v]dx ≤ Ω f Tk [u − v]dx k Ω

∀v ∈ Kψ .

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

175

It is our purpose, in this paper, to show the existence and uniqueness of solutions for the problem (3-6) in the setting of the Orlicz Sobolev space W01 LM (Ω). Our result, theorem 3.1, generalizes that of Boccardo8 and gives in particular a refinement of his result. Theorem 3.1. Assume that (3 − 1) − (3 − 5) hold true, and suppose that AM 6= ∅, then there exists an unique solution of problem (3 − 6). Proof of Theorem 3.1 see4

4. Existence of solutions for nonlinear elliptic degenerated equations Let Ω be a bounded open subset of IRN . A is a nonlinear operator of the Leray-lions type from a weighted Sobolev space W 1,p (Ω, ν) (where ν = ν(x) is weight function defined on Ω). A is defined by A(u) = −div a(x, u, ∇u) where a is caratheodory function from Ω × IR × IR N → IRN satisfying for a.e. x ∈ Ω, and for all ξ0 , ξ, (4 − 1)

1

1

| a(x, ξ0 , ξ) |≤ K(| ξ0 |)ν p (x){d(x)+ | ξ0 |p−1 +ν 1− p | ξ |p−1 },

where K verify that sup ess K(| s |) < ∞, |s|≤k

(4 − 2)

0

∀k > 0, and d(x) ∈ Lp (Ω).

[a(x, ξ0 , ξ) − a(x, ξ0 , η)] [ξ − η] > 0 where ξ 6= η.

Let λ be a continuous function such that λ > 0 and defined on IR+ , the degeneracy of the operator A is expressed by the assumption (4 − 3)

a(x, ξ0 , ξ) ξ ≥ ν(x) λ(| ξ0 |) | ξ |p ,

holds for all ξ0 , ξ and let λ1 such that Z s p0 (4 − 4) λ1 (s) = [λ(t)] p dt, s ≥ 0 0

with λ1 (∞) = ∞.

Furthermore, we shall assume that for some m0 > 0,

(4 − 5)

p0

inf (mp.(1− r ) [

m≥m0

For instance λ(t) = t We will suppose that (4 − 6)

−γp p0

inf

m≤t≤m+1

λ(t)]) > 0.

for t ≥ 1, 0 < γ < 1 (small). −1

ν ∈ L1loc (Ω), ν p−1 ∈ L1 (Ω),

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

176

ν −s ∈ L1 (Ω)

(4 − 7)

with s ∈ (

N 1 , ∞) ∩ [ , ∞). p p−1

We consider the nonlinear elliptic problem (4 − 8)

−div a(x, u, ∇u) − divφ(u) + g(x, u) = f,

(4 − 9)

u = 0 on ∂Ω,

where φ = (φ1 , ..., φN ) satisfy φ ∈ (C 0 (IR))N .

(4 − 10)

Note that no growth hypothesis is assumed on the function φ. Let g(x, t) be a caratheodory function such that for a.e.x ∈ Ω and all t ∈ IR, (4 − 11)

g(x, t) t ≥ 0 sup | g(., t) |= hn (.) ∈ L1 (Ω) ∀n.

(4 − 12)

|t|≤n

The right-hand side in (4-8) is of the form (4 − 13)

f=

N X ∂fi , ∂x i i=1

where the family {fi i = 1, ..., N } satisfies the conditions fi ∈ Lr (Ω, ν

−r p

0

) ,→ Lp (Ω, ν

−p0 p

) = (Lp (Ω, ν))∗ for i = 1, . . . , N, with r ≥ p0 .

We distinguish two cases. q N The first case if r > rc with rc = q−N . p−1 and q > N . In this case we prove in Theorem 4.1 that the solution u of problem (4-8)-(4-9) is bounded, and then the term divφ(u) has meaningful as a distribution. In Theorem 4.2, we prove that the last problem admits a weak solution (see Definition 4.1). The second case if p0 ≤ r ≤ rc the solution u is not bounded and the term divφ(u) may be meaningless, even as a distribution. As in6 we define the notion of renormalized solution (see Definition 4.2), which gives a meaning to a possible solution of (4-8)-(4-9). In Theorem 4.3 we prove the existence of renormalized solution. We show in the paper L∞ estimates for the solutions (see Theorem 4.1) and the existence of a weak and renormalized solution, and is also showed the fact that the main operator is degenerate in the space variable.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

177

4.1. Regularity and existence of weak solutions Definition 4.1. Assume that (4 − 1) − (4 − 7), (4 − 10) − (4 − 13) hold true.A function u is a weak solution of the problem (4 − 8) − (4 − 9) if u ∈ W01,p (Ω, ν) ∩ L∞ (Ω), g(x, u) ∈ L1 (Ω), u g(x, u) ∈ L1 (Ω)

(4 − 14) (4−15)

Z

Ω

a(x, u, ∇u)∇ϕdx+

Z

φ(u)∇ϕdx+ Ω

Z

g(x, u)ϕdx = Ω

N Z X ∂fi ϕdx ∂xi i=1 Ω

∀ϕ ∈ D(Ω). In Theorem 4.1 we prove the boundness of the solution of the problem (48)-(4-9), in Theorem 4.2 we prove the existence of weak solution if r > rc . Lemma 4.1. Let Ω be a bounded open subset of IR N . If u ∈ (W01,p (Ω, ν))N , then Z

div u dx = 0. Ω

Lemma 4.2. Let F : IR → IR be uniformly lipschitzian with F (0) = 0. Let u ∈ W01,p (Ω, ν), then F (u) ∈ W01,p (Ω, ν). Let u ∈ W01,p (Ω, ν) and let Sθ,h a real lipschitzian function defined for θ > 0, h > 0 by  1 if τ ≥ θ + h    τ −θ  if θ ≤ τ ≤ θ + h  h Sθ,h (τ ) = 0 if | τ |≤ θ  τ +θ   if −θ − h ≤ τ ≤ −θ   h −1 if τ ≤ −θ − h

then Sθ,h (u) ∈ W01,p (Ω, ν) by Lemma 4.2 and we suppose that u satisfies ∀θ ∈ ]0, sup ess | u | [, ∀h ∈ ]0, sup ess | u | −θ[, Z Z Z a(x, u, ∇u)∇Sθ,h (u)dx + φ(u)∇Sθ,h (u) + g(x, u)Sθ,h (u)dx Ω Ω Ω Z N (4−16) X ∂fi Sθ,h (u)dx = ∂xi i=1 Ω Remark 4.1. It is easy to see that each possible solution of problem (48)-(4-9) verifies the relation (4-16).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

178

Theorem 4.1. Under assumptions (4 − 1) − (4 − 7), (4 − 10) − (4 − 13). Let u ∈ W01,p (Ω, ν) which satisfies (4-16), and we suppose furthermore fi ∈ −r q N . p−1 and q > N . Then u is bounded Lr (Ω, ν p ), with r > rc where rc = q−N and we have the following estimate λ1 (k u k∞ ) ≤ CN α1 α2 α3 ⇔ k u k∞ ≤ λ−1 1 (CN α1 α2 α3 ) = M,

(4 − 17) where

−1

CN =

1

α1 =k ν p kLq (Ω) , PN 1 −r ) p−1 , α2 = ( i=1 k fi k r p ) L (Ω,ν R |Ω| 1 1 1 + α3 = ( 0 σ m(1− N ) dσ) m , with m 1

N N αN

1 r(p−1)

+

1 q

= 1,

, αN is the measure of the unit ball of IRN ,

λ−1 1 is the converse of the function λ1 (see (4-4)). Proof of Theorem 4.1. The proof is based on the method of relative rearrangement see5 .

Theorem 4.2. Under assumptions (4 − 1) − (4 − 7), (4 − 10) − (4 − 13). There exist a bounded weak solution u ∈ W01,p (Ω, ν) ∩ L∞ (Ω) (in the sense of Definition 4.1) of problem (4 − 8) − (4 − 9). Proof of Theorem 4.2. See5 .

4.2. Existence of renormalized solutions If p0 ≤ r ≤ rc , the solution of the problem (4-8)-(4-9) may not be bounded and the term div(φ(u)) may have no meaning even a distribution. Then we consider the nonlinear elliptic problem (4-8)-(4-9) in Ω, and we consider the following notion of renormalized solution which gives a meaning to a possible solution of (4-8)-(4-9). Definition 4.2. Assume that (4 − 1) − (4 − 7), (4 − 10) − (4 − 13) hold true. A function u is a renormalized solution of the problem (4 − 8) − (4 − 9) if u ∈ W01,p (Ω, ν), g(x, u) ∈ L1 (Ω), u g(x, u) ∈ L1 (Ω)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

179

Z  Z 0 0   Φ(u)[h (u)∇uϕ a(x, u, ∇u) [h (u)∇uϕ + h(u)∇ϕ]dx +    Ω  Ω Z N Z X ∂fi + h(u)∇ϕ]dx + g(x, u)h(u)ϕdx = h(u)ϕdx   ∂xi  Ω  i=1 Ω   ∀h ∈ Cc1 (IR), ϕ ∈ W01,p (Ω, ν) ∩ L∞ (Ω)

In Theorem 4.3 we state the existence of renormalized solution of (4-8)-(49). Theorem 4.3. Under assumptions (4 − 1) − (4 − 7), (4 − 10) − (4 − 13), N < r ≤ rc with q > N , there exists a renormalized and furthermore for p−1 solution u (in the sense of Definition 4.2) of problem (4 − 8) − (4 − 9) Proof of Theorem 4.3 is given in5 . References 1. L. Aharouch, J. Bennouna and A. Touzani, Existence of renormalized solution of some elliptic problems in Orlicz spaces, Revista Matematica Complutense, 22 (2009), n1, 91-110. 2. A. Benkirane, J. Bennouna, Existence of Renormalized Solutions for some elliptic problems involving derivatives of nonlinear terms in Orlicz Spaces, Lectures notes in pure and applied mathematics , 229, Dekker, New York, 2002. 3. A. Benkirane, J. Bennouna, Existence of Entropy Solutions for some nonlinear problems in L1 in Orlicz Spaces, Abstract and Applied Analysis 7 (2002) n 2, 85-102 USA. 4. A. Benkirane, J. Bennouna, Existence and uniqueness of solution of unilateral problems with L1 -data in Orlicz spaces, Ital. J. of Pure and App. Math., 16,(2004) 87-102. 5. A. Benkirane, J. Bennouna, Existence of Solutions for nonlinear elliptic degenerate equations, Nonlinear Analysis 54 (2003), n9-37, USA. 6. L. Boccardo, D. Giachetti, J.I. Diaz, F. Murat, Existence and Regularity of Renormalized Solutions of some Elliptic Problems involving derivatives of nonlinear terms, Journal of differential equations 106,215-237 (1993) 7. L. Boccardo, Some nonlinear Dirichlet problems in L1 involving lower order terms in divergence form, progress in Elliptic and Parabolic Partiel Differentiel Equations (capri, 1994), Pitman Res. Notes Math. Ser., vol. 350, Longman, Harlow, 1996, pp. 43-57. 8. L. Boccardo and G. R. Cirmi, Existence and uniqueness of solution of unilateral problems with L1 data, Journal of Convex Analysis, 6, N1 (1999) 195-206. 9. G. Dalmaso, F. Murat, L. Orsina and A. Prignet, Renormalized solutions of elliptic equations with general measure data, Ann. Scuola Norm. Sup Pisa Cl. Sci 12 4 (1999) 741-808.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

180

10. R.J. Diperna, P.L. Lions, on the cauchy problem for Boltzmann equations: Global existence and weak stability, ann of Math, 130 (1989), 321-366. 11. M. Kbiri Alaoui, Sur certains problems elliptiques avec second membre mesure ou L1loc , These soutenu a ` l’Universit´e Sidi Moh. Ben Abd. Facult´e des Sciences Dhar Mehraz Fes Maroc (1999) 12. J.M. Rakotoson, Uniqueness of renormalized solutions in a T -set for L 1 data problems and the link between various formulations, Indiana University Math. Jour., vol. 43, 2(1994). 13. G. Talenti, Nonlinear Elliptic Equations, Rearrangements of functions and Orlicz spaces, Ann. Mat. Pura Appl.(4) 120 (1979), 159-184.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

181

Existence of solutions for variational degenerated unilateral problems L. Aharouch, E. Azroul, and M. Rhoudaf D´ epartement de Math´ ematiques et Informatique Facult´ e des Sciences Dhar-Mahraz B.P 1796 Atlas F` es, Morocco An existence result is proved for a variational degenerated unilateral problems associated to the following equations Au + g(x, u, ∇u) = f, where A is a Leray-Lions operator acting from the weighted Sobolev space 0 W01,p (Ω, w) into its dual W −1,p (Ω, w ∗ ), while g(x, s, ξ) is a nonlinear term which has a growth condition with respect to ξ and a sign condition on s, i.e. g(x, s, ξ).s ≥ 0 for every s ∈ IR and for every x and ξ in their respective 0 domains. The source term f is supposed to belong to W −1,p (Ω, w ∗ ). Keywords: Degenerate unilateral problem; Existence result.

1. Introduction Let Ω be a bounded open subset of IRN (N ≥ 2), p be a real number such that 1 < p < ∞ and w = {wi (x); 0 ≤ i ≤ N } , be a collections of weight functions on Ω, i.e. each wi (x) is a measurable a.e. strictly positive function on Ω satisfying some intergrability conditions (see section 2). In this paper we are interested in the study of the degenerated obstacle problem associated to the following Dirichlet problem Au + g(x, u, ∇u) = f in Ω (1) u ≡ 0 on ∂Ω, where Au = −div(a(x, u, ∇u)) is a Leray-Lions operator act−1,p0 ing from W01,p (Ω, (Ω, w∗ ) with w∗ = n o w) into its dual W 0 p wi1−p ; 0 ≤ i ≤ N , p0 = p−1 is the conjugate exponent of p and where

g(x, u, ∇u) is a nonlinearity term satisfying some p-growth condition with respect to ∇u, and satisfies the sign condition g(x, u, ∇u)u ≥ 0, but has unrestricted growth with respect to u.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

182 0

The source term f is assumed to belong to W −1,p (Ω, w∗ ). Let us start by the case of equation and recall that in the particular case where g(x, u, ∇u) = −C0 |u|p−2 u the following degenerated equation −div(a(x, u, ∇u)) − C0 |u|p−2 u = h(x, u, ∇u), has been studied by Drabek-Nicolosi9 under some more degeneracy and some additional assumptions on h and a. While the existence solution for the variational Dirichlet problem (1) is treated in the work3 but under the following integrability condition σ −1/q−1 ∈ L1loc (Ω) for some q such that 1 < q < ∞,

(2)

where q is the so-called Hardy exponent and σ is the so-called Hardy weight (see (14)) below). Now we turn our attention to the degenerated unilateral case and we will give some known about the following problem  Find u ∈ Kψ , g(x, u, ∇u) ∈ LZ1 (Ω), g(x, u, ∇u)u ∈ L1 (Ω)   Z    a(x, u, ∇u)∇(u − v) dx + g(x, u, ∇u)(u − v) dx (3) Ω Ω   ≤ hf, u − vi,    ∀ v ∈ Kψ ∩ L∞ (Ω), where the convex set Kψ is defined as

Kψ = {v ∈ W01,p (Ω, w); v ≥ ψ a.e. in Ω}, with an obstacle ψ which is a measurable function on Ω. Akdim et al. have proved in4 the existence of a solution for the problem (3) under the following conditions 1

σ − q−1 ∈ L1loc (Ω) with 1 < q < p + p0 ,

(4)

ψ + ∈ L∞ (Ω) ∩ W01,p (Ω, w).

(5)

and

For that, the authors in4 have approximated the nonlinear term g by some function involving χΩε where Ωε is a sequence of compacts covering the bounded open set Ω and χΩε is a characteristic function, i.e. gε (x, s, ξ) =

g(x, s, ξ) χΩ (x). 1 + ε|g(x, s, ξ)| ε

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

183

So there are consider the following approximate unilateral problem  uε ∈ Kψ such that   Z Z    a(x, u , ∇u )∇(u − v) dx + gε (x, uε , ∇uε )(uε − v) dx ε ε ε Ω Ω   ≤ hf, uε − vi,    ∀ v ∈ Kψ ,

(6)

Note that, the hypotheses (4) and (5) used in4 have played an important role for to assure the boundedness, coercivity and pseudo-monotonicity of the operator associated to the approximate problem (6) and also for to prove the boundedness of the approximate solution uε in the space W01,p (Ω, w) (see3 for more details). The aim of this paper is then to study the existence solution for the same degenerated unilateral problem (3) but without assuming the condition (4) nor (5). To overcome the difficulties mentioned above, we have changed in the present paper the classical coercivity, a(x, s, ξ)ξ ≥ α by the following one,

N X i=1

a(x, s, ξ)(ξ − ∇v0 ) ≥ α

wi (x)|ξi |p

N X i=1

wi (x)|ξi |p − δ(x),

(7)

where v0 is some element of Kψ ∩ L∞ (Ω) and δ(x) is an element of L1 (Ω) and also we have replaced the approximation term gε (of the nonlinearity g) by the following one, gn (x, s, ξ) =

g(x, s, ξ) θn (x), 1 + n1 |g(x, s, ξ)|

(8)

where the function θn (x) is defined according to the Hardy weight σ and hardy exponent q, i.e., θn (x) = nT1/n (σ 1/q (x)). (T1/n is the truncation operator at height 1/n see (18)). It would be interesting at this stage to refer the reader to our previous work1 in which the same unilateral problem with L1 -data is studied under the integrability condition (4) and under the regularity condition (5). Another work in this direction can be found in2 where the existence solution of the same unilateral problem (with L1 -data) is proved by assuming the

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

184

previous hypotheses (4) and (5), but the sign condition (22) is violated and the classical growth condition of the nonlinearity g (see (23)) is replaced by |g(x, s, ξ)| ≤ c(x) + ρ(s)

N X i=1

wi |ξi |p ,

(9)

(with ρ ∈ L1 (Ω), ρ ≥ 0). We refer also to the work,14 where the authors have solved an analogous problem in the classical Sobolev space but where the obstacle function is supposed to satisfy the condition ψ + ∈ W01,p (Ω) ∩ L∞ (Ω). The outline of this paper is as follows. After giving some preliminary results about the weighted Sobolev space in section 2, we formulate in section 3 our problem and we give the main existence result, which its proof is giving in section 4. And we achieve the paper by an appendix in section 5 where some intermediate results are proved. 2. Preliminaries Let Ω be a bounded open subset of IRN (N ≥ 2). Let 1 < p < ∞ and let w = {wi (x); i = 0, . . . , N } , be a vector of weight functions, i.e., every component wi (x) is a measurable function which is strictly positive a.e. in Ω. Further, we suppose in all our considerations that, wi ∈ L1loc (Ω)

(10)

and 1 − p−1

wi

∈ L1loc (Ω), for 0 ≤ i ≤ N.

(11)

We define the weighted space with weight γ in Ω as, 1

Lp (Ω, γ) = {u(x), uγ p ∈ Lp (Ω)}, which is normed by, kukp,γ =

Z

p

Ω

|u(x)| γ(x) dx

p1

.

We denote by W 1,p (Ω, w) the weighted Sobolev space of all real-valued functions u ∈ Lp (Ω, w0 ) such that the derivatives in the sense of distributions satisfy, ∂u ∈ Lp (Ω, wi ) for all i = 1, . . . , N. ∂xi

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

185

This set of functions forms a Banach space under the norm Z

kuk1,p,w =

Ω

|u(x)|p w0 dx +

N Z X i=1

∂u p | | wi (x) dx ∂x i Ω

! p1

.

(12)

To deal with the Dirichlet problem, we use the space W01,p (Ω, w) defined as the closure of C0∞ (Ω) with respect to the norm (12). Note that C0∞ (Ω) is dense in W01,p (Ω, w) and (W01,p (Ω, w), k.k1,p,w ) is a reflexive Banach space. We recall that the dual of the weighted Sobolev spaces W01,p (Ω, w) is 0 0 equivalent to W −1,p (Ω, w∗ ), where w∗ = {wi∗ = wi1−p }, i = 1, . . . , N and p . For more details we refer the reader p0 is the conjugate of p, i.e., p0 = p−1 to.11 Now, we state the following assumptions. (H1 ) The expression, kuk =

N Z X i=1

∂u p | wi (x) dx | ∂x i Ω

! p1

(13)

is a norm on W01,p (Ω, w) equivalent to the norm (12). There exists a weight function σ strictly positive a.e. in Ω and a parameter q, 1 < q < ∞, such that the Hardy inequality ! p1 Z 1q N Z X ∂u p q | wi (x) dx , (14) ≤C | |u| σ(x) dx ∂xi Ω i=1 Ω holds for every u ∈ W01,p (Ω, w) with a constant C > 0 independent of u. Moreover, the imbedding W01,p (Ω, w) ,→ Lq (Ω, σ)

(15)

determined by the inequality (14) is compact. Note that, (W01,p (Ω, w), k k) is a uniformly convex (and thus reflexive) Banach space. Remark 2.1. Assume that w0 (x) = 1 and in addition the integrability condition: 1 , ∞[ such that wi−ν ∈ L1 (Ω) holds for all There exists ν ∈] Np , ∞[∩[ p−1 i = 1, . . . , N (which is stronger than (11)). Then ! p1 N Z X ∂u p | kuk = | wi (x) dx ∂xi i=1 Ω

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

186

is a norm defined on W01,p (Ω, w) and it is equivalent to (12). Moreover W01,p (Ω, w) ,→,→ Lq (Ω) for all 1 ≤ q < p∗1 if pν < N (ν + 1) and for all q ≥ 1 if pν ≥ N (ν + 1), pν p1 N pν where p1 = ν+1 and p∗1 = NN−p = N (ν+1)−pν is the Sobolev conjugate of 1 11 p1 (see ). Thus the hypotheses (H1 ) is satisfied for σ ≡ 1. Remark 2.2. If we use the special weight functions w and σ expressed in terms of the distance to the boundary ∂Ω. Denote d(x) = dist(x, ∂Ω) and set w(x) = dλ (x),

σ(x) = dµ (x).

In this case, the Hardy inequality reads Z q1 Z p1 q µ p λ |u| d (x) dx ≤C . |∇u| d (x) dx Ω

Ω

(i) For, 1 0. q p q p

(16)

(ii) For, 1 ≤ q 0. q p q p

(17)

The conditions (16) or (17) are sufficient for the compact imbedding (15) to hold (see for example [,10 Example 1], [,11 Example 1.5, p.34], and [,16 theorems 19.17 and 19.22]). Now, we give the following technical lemmas which are needed later. Lemma 2.1. cf.3,15 Let g ∈ Lr (Ω, γ) and let gn ∈ Lr (Ω, γ), with kgn kΩ,γ ≤ c, 1 < r < ∞. If gn (x) → g(x) a.e. in Ω, then gn * g weakly in Lr (Ω, γ). Lemma 2.2. cf.3,12 Assume that (H1 ) holds. Let F : IR → IR be uniformly Lipschitzian, with F (0) = 0. Let u ∈ W01,p (Ω, w). Then F (u) ∈ W01,p (Ω, w). Moreover, if the set D of discontinuity points of F 0 is finite, then 0 ∂u ∂(F ◦ u) F (u) ∂x a.e. in {x ∈ Ω : u(x) ∈ / D} i = ∂xi 0 a.e. in {x ∈ Ω : u(x) ∈ D}.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

187

We introduce the truncation operator. For a given constant k > 0 we define the cut function Tk : IR → IR as s if |s| ≤ k Tk (s) = (18) k sign(s) if |s| > k. For a function u = u(x), x ∈ Ω, we define the truncated function Tk u = Tk (u) pointwise: for every x ∈ Ω the value of (Tk u) at x is just Tk (u(x)). From Lemma 2.2, we deduce the following. Lemma 2.3. cf.3 Assume that (H1 ) holds. Let u ∈ W01,p (Ω, w) and let Tk (u) be the usual truncation (k ∈ IR+ ). Then Tk (u) ∈ W01,p (Ω, w). Moreover, we have Tk (u) → u strongly in W01,p (Ω, w). 0

We states that every element of W −1,p (Ω, w∗ ) can be decomposed as f0 − N Y 0 0 0 0 divF where f0 ∈ Lp (Ω, w01−p ) and F ∈ Lp (Ω, wi1−p ). i=1

3. Main results Let Ω be an open bounded subset of IRN , N ≥ 2. Given a measurable function ψ : Ω → IR (so-called obstacle) and consider its associated convex set, Kψ = {u ∈ W01,p (Ω, w); u ≥ ψ a.e. in Ω}.

(19)

Let A be the nonlinear operator from W01,p (Ω, w) into its dual 0 W −1,p (Ω, w∗ ) defined by, Au = −div(a(x, u, ∇u)), where a : Ω × IR × IRN → IRN is a Carath´eodory function satisfying the following assumptions: (H2 ) For i = 1, . . . , N 1

1

q

|ai (x, s, ξ)| ≤ wip (x)[k(x) + σ p0 |s| p0 +

N X j=1

1 0

wjp (x)|ξj |p−1 ],

[a(x, s, ξ) − a(x, s, η)](ξ − η) > 0 for all ξ 6= η ∈ IR N ,

(20)

(21)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

188 0

for a.e. x in Ω, s ∈ IR, where k(x) is a positive function in Lp (Ω). (H3 ) g(x, s, ξ) is a Carath´eodory function which satisfies the following classical conditions, g(x, s, ξ).s ≥ 0

(22)

and |g(x, s, ξ)| ≤ b(|s|)(

N X i=1

wi (x)|ξi |p + c(x)),

(23)

where b : IR+ → IR+ is a positive increasing function and c(x) is a positive function in L1 (Ω). Our main result in this note is the following 0

Theorem 3.1. Assume that (H1 )−(H3 ) and (7) hold and f ∈ W −1,p (Ω). Then there exists at least one solution of the following unilateral problem,  1 1 u   Z ∈ Kψ g(x, u, ∇u) ∈ L (Ω), Zg(x, u, ∇u)u ∈ L (Ω)    a(x, u, ∇u)∇(u − v) dx + g(x, u, ∇u)(u − v) dx (24) Ω Ω   ≤ hf, u − vi,    ∀ v ∈ Kψ ∩ L∞ (Ω). Remark 3.1. Note that if we take w ≡ 1 and σ ≡ 1 in the statement of the previous theorem, then we get a new result in the nondegenerated case.

4. Proof of Theorem 3.1 For the reason of simplicity we take f N Y 0 0 Lp (Ω, wi1−p ).

=

−divF where F

∈

i=1

The following lemma play an important rˆ ole in the proof of our main result,

Lemma 4.1. 4,15 Assume that (H1 ), (H2 ) and (7) are satisfied, and let (un )n be a sequence in W01,p (Ω, w) such that un * u weakly in W01,p (Ω, w) and Z [a(x, un , ∇un ) − a(x, un , ∇u)]∇(un − u) dx → 0 Ω

then, un → u in W01,p (Ω, w).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

189

We shall give the proof of Theorem in several steps. STEP 1: Construction of the approximate unilateral problem and existence of a solution. Let us approximate the nonlinear function g by gn defined through (8). We consider the following approximate unilateral problem  u ∈ Kψ ,   Z  n hAun , un − vi + gn (x, un , ∇un )(un − v) dx ≤ hf, un − vi (25)  Ω   ∀v ∈ Kψ . It is easy prove that gn satisfies the following conditions:

gn (x, s, ξ)s ≥ 0, |gn (x, s, ξ)| ≤ |g(x, s, ξ)| and |gn (x, s, ξ)| ≤ n for a.e. x in Ω, s ∈ IR and all ξ ∈ IRN . 0 Thus, we can define the operator Gn : W01,p (Ω, w) −→ W −1,p (Ω, w∗ ) by,

hGn u, vi = and hAu, vi =

Z

Z

Ω

Ω

gn (x, u, ∇u)v dx

a(x, u, ∇u)∇v dx.

By virtue of H¨ older’s inequality and due to (13) and (14), we have for all 1,p u ∈ W0 (Ω, w) and all v ∈ W01,p (Ω, w), Z Z q1 10 Z 0 q q 0 − qq q gn (x, u, ∇u)v dx ≤ |gn (x, u, ∇u)| σ dx |v| σ dx Ω Ω Ω Z 10 q 0 0 kvkq,σ ≤n σ q /q σ −q /q dx Ω

≤ Cn kvk.

(26)

Definition 4.1. Let Y be a reflexive Banach space. A bounded operator B from Y to its dual Y ∗ is called pseudo-monotone if for any sequence un ∈ Y with un * u weakly in Y , and lim suphBun , un − ui ≤ 0, we have n→+∞

lim inf hBun , un − vi ≥ hBu, u − vi ∀ v ∈ Y. n→+∞

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

190 0

Consider the operator Bn : W01,p (Ω, w) −→ W −1,p (Ω, w∗ ) defined as hBn v, wi =

Z

Ω

gn (x, v, ∇v)w dx +

Z

Ω

a(x, v, ∇v)∇w dx.

(27)

0

Lemma 4.2. The operator Bn from Kψ into W −1,p (Ω, w∗ ) is pseudomonotone. Moreover, Bn is coercive in the following sense: hBn v, v − v0 i −→ +∞ if kvk −→ +∞, v ∈ Kψ , kvk where v0 is associated to a through (7). The proof of this Lemma will be given in Appendix 1. Lemma 4.2 implies that the problem (25) has a solution, applying the classical result for nonlinear equations associated to pseudo-monotone operators (see [,15 Theorem 8.2]). STEP 2: A priori estimates on the approximate solution and convergence. 2 2 Let k ≥ kv0 k∞ and let ϕk (s) = seγs , where γ = ( 2b(k) α ) . An essential role will be played the following property enjoyed by ϕk (s) (the proof is trivial): ϕ0k (s) −

1 2b(k) |ϕk (s)| ≥ , ∀s ∈ IR. α 2

(28)

Taking un − ηϕk (Tl (un − v0 )) (η small enough) as test function in (25), where l = k + kv0 k∞ , we obtain Z a(x, un , ∇un )∇Tl (un − v0 )ϕ0k (Tl (un − v0 )) dx Ω Z + gn (x, un , ∇un )ϕk (Tl (un − v0 )) dx ≤ hf, ϕk (Tl (un − v0 ))i. Ω

Using the fact that gn (x, un , ∇un )ϕk (Tl (un − v0 )) ≥ 0 on the subset {x ∈ Ω : |un (x)| > k}, we get Z a(x, un , ∇un )∇(un − v0 )ϕ0k (Tl (un − v0 )) dx {|un −vZ 0 |≤l} ≤ |gn (x, un , ∇un )||ϕk (Tl (un − v0 ))| dx Z {|un |≤k} + F ∇Tl (un − v0 )ϕ0k (Tl (un − v0 )) dx. Ω

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

191

Now we use (7) and (23), we can write α

Z

N X

∂un p 0 | ϕk (Tl (un − v0 )) dx ∂xi {|un −v0 |≤l} i=1 ! Z N X ∂Tk (un ) p wi (x)| | |ϕk (Tl (un − v0 ))| dx ≤ b(|k|) c(x) + ∂xi Ω i=1 Z Z F.∇un ϕ0k (Tl (un − v0 )) dx + δ(x)ϕ0k (Tl (un − v0 )) dx + {|u −v |≤l} Ω n 0 Z + F.∇v0 ϕ0k (Tl (un − v0 )) dx. wi (x)|

{|un −v0 |≤l}

Applying the Young’s inequality and the fact that F ∈ we get α

Z

N X

N Y

0

0

Lp (Ω, wi1−p ),

i=1

∂un p 0 | ϕk (Tl (un − v0 )) dx ∂xi {|un −v0 |≤l} i=1 ! Z N X ∂Tk (un ) p ≤ b(|k|) c(x) + wi (x)| | |ϕk (Tl (un − v0 ))| dx ∂xi Ω i=1 Z N X ∂un p 0 α +2 wi (x)| | ϕk (Tl (un − v0 )) dx ∂xi Z {|un −v0 |≤l} i=1 + δ(x)ϕ0k (Tl (un − v0 )) + C1 (k) wi (x)|

Ω

where C1 (k) is a positive constant depending on k. Thanks to {x ∈ Ω, |un (x)| ≤ k} ⊆ {x ∈ Ω : |un − v0 | ≤ l} and the fact that c, δ ∈ L1 (Ω), we have Z X N Ω i=1

wi (x)|

∂Tk (un ) p 0 2b(k) |ϕk (Tl (un −v0 ))| dx ≤ C2 (k). | ϕk (Tl (un −v0 ))− ∂xi α

Thus by (28), we deduce Z X N

wi (x)|

Ω i=1

∂Tk (un ) p | dx ≤ 2C2 (k). ∂xi

Now, we claim that, Z X N Ω i=1

wi (x)|

∂un p | dx ≤ C. ∂xi

(29)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

192

Let k ≥ kv0 k∞ . Taking v = v0 as a test function in (25), we get Z Z gn (x, un , ∇un )(un − v0 ) dx a(x, un , ∇un )∇(un − v0 ) dx + Z Ω Ω ≤ F ∇(un − v0 ) dx.

(30)

Ω

Furthermore, since gn (x, un , ∇un )(un − v0 ) ≥ 0 on the subset {x ∈ Ω, |un (x)| > k}, the inequality (30) implies that Z a(x, un , ∇un )∇(un − v0 ) dx Z Z Ω ≤ 2k |gn (x, un , ∇un )| dx + F ∇(un − v0 ) dx, {|un |≤k}

Ω

which gives by using (23) and Young’s inequality Z a(x, un , ∇un )∇(un − v0 ) dx Ω "Z # Z X N ∂Tk (un ) p ≤ 2kb(k) c(x) dx + wi (x)| | dx ∂xi Ω Ω i=1 Z X N ∂un p wi (x)| | dx + C. + α2 ∂xi Ω i=1 (31) Using (29) and (31), we obtain Z Z X N ∂un p α a(x, un , ∇un )∇(un − v0 ) dx ≤ 2 wi (x)| | dx + C3 (k). ∂xi Ω Ω i=1

Thanks to (7), we have

Z X N Ω i=1

wi (x)|

∂un p | dx ≤ C. ∂xi

(32)

Then, we conclude that there exists a function u ∈ W01,p (Ω, w) such that un * u weakly in W01,p (Ω, w), un → u strongly in Lq (Ω, σ) and a.e. in Ω.

This yields, by using (20), the existence of a function h ∈ such that a(x, un , ∇un ) * h weakly in

N Y

i=1

0

N Y

(33) 0

0

Lp (Ω, wi1−p ),

i=1

0

Lp (Ω, wi1−p ).

(34)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

193

STEP 3: Almost everywhere convergence of the gradient. We fix k > kv0 k∞ , and let wn,k = Tk (un ) − Tk (u). For η > 0, we consider the following function: vn = un − ηϕk (wn,k ).

(35)

We choose η such that vn ∈ Kψ (see Appendix 2). As in,7 we take vn as a test functions in (25), we get Z Z gn (x, un , ∇un )ϕk (wn,k ) dx a(x, un , ∇un )∇wn,k ϕ0k (wn,k ) dx + (36) Ω Ω ≤ hf, ϕk (wn,k )i. Remark that, we have  if un ≥ k  k − Tk (u) ≥ 0 Tk (un ) − Tk (u) = un − Tk (u) ≤ 0 if |un | ≤ k  −k − Tk (u) ≤ 0 if un ≤ −k

which implies, by using (22), g(x, un , ∇un )ϕk (wn,k ) ≥ 0 on the set {x ∈ Ω, |un (x)| > k}. So by (36), Z Z 0 a(x, un , ∇un )∇wn,h ϕk (wn,k ) dx + gn (x, un , ∇un )ϕk (wn,k ) dx Ω

{|un |≤k}

≤ hf, ϕk (wn,k )i.

(37) Splitting the first integral on the left hand side of (37) where |un | ≤ k and |un | > k, we can write, Z a(x, un , ∇un )∇wn,k ϕ0k (wn,k ) dx Z Ω = a(x, un , ∇un )[∇Tk (un ) − ∇Tk (u)]ϕ0k (wn,k ) dx {|u |≤k} n Z + a(x, un , ∇un )∇wn,k ϕ0k (wn,k ) dx. {|un |>k

(38) Now we estimate the first term in the right hand side of (38) as follows Z a(x, un , ∇un )∇wn,k ϕ0k (wn,k ) dx {|un |≤k} Z ≥ a(x, Tk (un ), ∇Tk (un ))[∇Tk (un ) − ∇Tk (u)]ϕ0k (wn,k ) dx Ω Z N X ∂Tk (u) |ai (x, Tk (un ), 0)|| − ϕ0k (2k) | dx. ∂xi {|un |>k} i=1 (39)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

194

We claim that last term of the above inequality goes to zero as n tends to zero. Indeed we have, for that i = 1, . . . , N |ai (x, Tk (un ), 0)|χ{|un |>k} con0

0

verges to |a(x, Tk (u), 0)|χ{|u|>k} strongly in Lp (Ω, wi1−p ), moreover, since k (u) | ∂T∂x | ∈ Lp (Ω, wi ), then i −ϕ0k (2k)

Z

N X

{|un |>k} i=1

|ai (x, Tk (un ), 0)||

∂Tk (u) | dx = ε(n). ∂xi

Where ε(n) is a quantity (which is possible to be changed from a line to another) such that lim ε(n) = 0 We now investigate the behaviour of the n→+∞

second term in the right hand side of (38) we get, Z

{|un |>k}

a(x, un , ∇un )∇wn,k ϕ0k (wn,k ) dx ≥

−ϕ0k (2k)

Z

N X

{|un |>k} i=1

|ai (x, un , ∇un )||

∂Tk (u) | dx. ∂xi (40)

By (34) and the fact that ∂Tk (u) ∂Tk (u) χ{|un |>k} → χ{|u|>k} = 0 ∂xi ∂xi in Lp (Ω, wi ), we have by (34)

−ϕ0k (2k)

Z

N X

{|un |>k} i=1

|ai (x, un , ∇un )||

∂Tk (u) | dx = ε(n) ∂xi

(41)

Gathering (39), (40) and (41) we have that (38) implies: Z

Ω

a(x, un , ∇un )∇wn,k ϕ0k (wn,k ) dx Z ≥ a(x, Tk (un ), ∇Tk (un ))[∇Tk (un ) − ∇Tk (u)]ϕ0k (wn,k ) dx Ω

+ ε(n).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

195

Which implies that Z a(x, un , ∇un )∇wn,k ϕ0k (wn,k ) dx Ω Z ≥ [a(x, Tk (un ), ∇Tk (un )) − a(x, Tk (un ), ∇Tk (u))] Ω

0 ×[∇T Z k (un ) − ∇Tk (u)]ϕk (wn,k ) dx + a(x, Tk (un ), ∇Tk (u))[∇Tk (un ) − ∇Tk (u)]ϕ0k (wn,k ) dx Ω

+ ε(n).

(42) By the continuity of the Nemitskii operator (see ), we have for all i = 1, . . . , N, 11

ai (x, Tk (un ), ∇Tk (u))ϕ0 (Tk (un ) − Tk (u)) → ai (x, Tk (u), ∇Tk (u))ϕ0 (0) 0

0

k (u)) k (un )) * ∂(T∂x weakly in Lp (Ω, wi ). strongly in Lp (Ω, wi1−p ), while ∂(T∂x i i Which yields Z a(x, Tk (un ), ∇Tk (u))[∇Tk (un ) − ∇Tk (u))]ϕ0k (wn,k ) dx = 0. (43) lim

n→∞

Ω

Du to (42) and (43), we conclude that Z

a(x, un , ∇un )[∇Tk (un ) − ∇Tk (u)]ϕ0 (wn,k ) dx Ω Z ≥ [a(x, Tk (un ), ∇Tk (un )) − a(x, Tk (un ), ∇Tk (u))] Ω

×[∇Tk (un ) − ∇Tk (u)]ϕ0 (wn,k ) dx + ε(n). (44) We now, discus the behaviour of the second integral of the left hand side of (37). Using (23), we have Z gn (x, un , ∇un )ϕk (wn,k ) dx {|un |≤k} Z N X ∂Tk (un ) p ≤ b(k) (c(x) + wi | | |ϕk (wn,k )| dx ∂xi i=1 ZΩ Z b(k) ≤ b(k) c(x)|ϕk (wn,k )| dx + α δ(x)|ϕk (wn,k )| ZΩ Ω + b(k) a(x, Tk (un ), ∇Tk (un ))∇Tk (un )|ϕk (wn,k )| dx α ZΩ − b(k) a(x, Tk (un ), ∇Tk (un ))∇v0 |ϕk (wn,k )| dx. α Ω

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

196

From (20) and (29), there exists a function hk ∈

N Y

i=1

a(x, Tk (un ), ∇Tk (un )) * hk weakly in

Moreover, since ∇v0 |ϕk (wn,k )| → 0 in

N Y

0

0

Lp (Ω, wi1−p ) such that

N Y

0

0

Lp (Ω, wi1−p )

(45)

i=1

Lp (Ω, wi ) as n tend to infinity,

i=1

the third term of the right hand side of the last inequality tend to zero as n tend to infinity, hence Z gn (x, un , ∇un )ϕk (wn,k ) dx {|un |≤k} Z ≤ b(k) a(x, Tk (un ), ∇Tk (un ))∇Tk (un )|ϕk (wn,k )| dx α ZΩ Z δ(x)|ϕk (wn,k )| dx + ε(n). + b(k) c(x)|ϕk (wn,k | dx + b(k) α Ω

Ω

(46) Next we estimate the first term in the right hand side of (46) as follows: Z b(k) a(x, Tk (un ), ∇Tk (un ))∇Tk (un )|ϕk (wn,k )| dx α Ω Z = b(k) [a(x, Tk (un ), ∇Tk (un )) − a(x, Tk (un ), ∇Tk (u)] α Ω

×[∇Tk (un ) − ∇Tk (u)]|ϕk (wn,k )| dx Z b(k) + α a(x, Tk (un ), ∇Tk (u))[∇Tk (un ) − ∇Tk (u))]|ϕk (wn,k )| dx ZΩ a(x, Tk (un ), ∇Tk (un ))∇Tk (u)|ϕk (wn,k )| dx. + b(k) α Ω

(47)

By Lebesgue’s Theorem, we deduce that ∇Tk (u)|ϕk (wn,k )| → ∇Tk (u)|ϕk (0)| = 0 strongly in

N Y

Lp (Ω, wi ).

i=1

Which and using (45) implies that the third term of (47) tends to 0 as n → ∞. On the other side reasoning as in (43), the second term of (47) tends to 0 as n → ∞.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

197

From (46) and (47), we obtain Z gn (x, un , ∇un )ϕk (wn,k ) dx {|un |≤k} Z ≤ [a(x, Tk (un ), ∇Tk (un )) − a(x, Tk (un ), ∇Tk (u))] Ω

×[∇TkZ(un ) − ∇Tk (u)]|ϕk (wn,k )| dx Z +b(k)

c(x)|ϕk (wn,k )| dx +

Ω

b(k) α

(48)

δ(x)|ϕk (wn,k )| dx + ε(n).

Ω

Combining (37), (44) and (48), we obtain Z [a(x, Tk (un ), ∇Tk (un )) − a(x, Tk (un ), ∇Tk (u))] Ω

×[∇TkZ(un ) − ∇Tk (u)](ϕ0k (wn,k ) −Z b(k) α |ϕk (wn,k )|) dx

≤ b(k)

c(x)|ϕk (wn,k )| dx +

Ω

+hf, ϕk (wn,k )i + ε(n).

b(k) α

δ(x)|ϕk (wn,k )| dx

Ω

(49) Therefore (28) implies Z [a(x, Tk (un ), ∇Tk (un )) − a(x, Tk (un ), ∇Tk (u))] Ω

Z ×[∇Tk (un ) − ∇Tk (u)] dx Z ≤ 2b(k) c(x)|ϕk (wn,k )| dx + 2 b(k) δ(x)|ϕk (wn,k )| dx α Ω

+hf, ϕk (wn,k )i + ε(n).

Ω

(50) Now, since c, δ ∈ L1 (Ω)), ϕk (wn,k ) * 0 weakly in W01,p (Ω, w) as n → ∞, all the terms of the right hand side of the last inequality tends to 0 as n → +∞. This implies that Z lim [a(x, Tk (un ), ∇Tk (un )) − a(x, Tk (un ), ∇Tk (u))][∇Tk (un ) n→∞ Ω

−∇Tk (u)] dx = 0.

Finally, Lemma 4.1 implies that Tk (un ) → Tk (u) strongly in W01,p (Ω, w) ∀k > 0.

(51)

Since k arbitrary, we have for a subsequence ∇un → ∇u a.e. in Ω.

(52)

Indeed, we show that (as in8 ) ∇un converge to ∇u in measure (and therefore, we can always assume that the convergence is a.e. after passing to a

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

198

suitable subsequence). Let k > 0 large enough, we have Z Z kmeas({|un | > k} ∩ BR ) = |Tk (un )| dx ≤ |Tk (un )| dx {|u |>k}∩BR BR p1 Z 10 Z n q 1−p0 p w0 |un | w0 dx dx ≤ BR Ω ! p1 Z X N ∂un p | ≤ c0 | wi (x) dx Ω i=1 ∂xi ≤ c1

where BR = {x ∈ Ω; |x| ≤ R}. Which implies that meas({|un | > k} ∩ BR ) ≤ same, since u ∈ W01,p (Ω, w), we get meas({|u| > k} ∩ BR ) ≤

c1 ∀k > 1. k

(53)

c2 ∀k > 1. k

(54)

We have, for every δ > 0, meas({|∇un − ∇u| > δ}) ≤ meas({|un | > k}) + meas({|u| > k}) +meas({|∇Tk (un ) − ∇Tk (u)| > δ}) ≤ meas({|un | > k} ∩ BR ) + meas({|u| > k} ∩ BR ) +2meas({|x| > R}) + meas({|∇Tk (un ) − ∇Tk (u)| > δ}). (55) Since Tk (un ) converge strongly in W01,p (Ω, w), we can assume that ∇Tk (un ) converge to ∇Tk (u) in measure in Ω. Let ε > 0, for R large enough, by (53), (54) and (55), there exists some n0 (k, R, δ, ε) > 0 such that meas({|∇un − ∇u| > δ}) < ε for all n, m ≥ n0 (k, R, δ, ε). This concludes the proof of (52). Which yields a(x, un , ∇un ) → a(x, u, ∇u) a.e. in Ω (56) gn (x, un , ∇un ) → g(x, u, ∇u) a.e. in Ω. Step 4: Equi-integrability of the nonlinearities. We need to prove that gn (x, un , ∇un ) → g(x, u, ∇u) strongly in L1 (Ω),

(57)

in particular it is enough to prove the equi-integrable of gn (x, un , ∇un ). To this purpose. We consider the function T1 (un − v0 − Th (un − v0 )) (with

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

199

h ≥ kv0 k∞ ). This function can be write as follows  0 if |un − v0 | ≤ h      un − v0 − h if h ≤ un − v0 ≤ h + 1 T1 (un − v0 − Th (un − v0 )) = un − v0 + h if −h − 1 ≤ un − v0 ≤ −h    1 if un − v0 ≥ h + 1   −1 if un − v0 ≤ −h − 1 which implies that

gn (x, un , ∇un )T1 (un − v0 − Th (un − v0 )) ≥ 0

(58)

and gn (x, un , ∇un )T1 (un − v0 − Th (un − v0 )) = |gn (x, un , ∇un )|

(59)

on the set {x ∈ Ω; |un − v0 | ≥ h + 1}. Now, we take un − T1 (un − v0 − Th (un − v0 )) as a test function in (25), we obtain by using (58) and (59) Z a(x, un , ∇un )∇(un − v0 ) dx {h≤|un −v0 |≤h+1} Z + |gn (x, un , ∇un )| dx Z{|un −v0 |>h+1} ≤ F ∇(un − v0 ) dx {h≤|un −v0 |≤h+1}

using the Young’s inequality and (7), we deduce Z N Z X 0 0 |gn (x, un , ∇un )| dx ≤ c |Fi |p wi1−p dx {|un −v0 |>h+1} i=1 {|un −v0 |>h} Z + (|F ∇v0 | + δ(x)) dx {|un −v0 |>h}

(60) 0 0 Let ε > 0, since the functions |Fi |p wi1−p , |F ∇v0 | and δ belongs to L1 (Ω) then there exists h(ε) ≥ 1 such that Z |g(x, un , ∇un )| dx < ε/2. (61) {|un −v0 |>h(ε)}

For any measurable subset E ⊂ Ω, we have Z Z |gn (x, un , ∇un )| dx ≤ b(h(ε) + kv0 k∞ )(c(x) E E Z N X ∂T (un ) p 0 k∞ | ) dx + |g(x, un , ∇un )| dx. + wi | h(ε)+kv ∂xi i=1

{|un −v0 |>h(ε)}

(62)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

200

In view of (51) there exists η(ε) > 0 such that Z

E

b(h(ε) + kv0 k∞ )(c(x) +

N X i=1

∂Th(ε)+kv0 k∞ (un ) p | ) dx < ε/2 ∂xi

wi |

(63)

for all E such that meas E < η(ε). Finally, combining (61), (62) and (63), one easily has Z |gn (x, un , ∇un )| dx < ε for all E such that meas E < η(ε), E

which implies (57) Step 5: Passing to the limit. We take v ∈ Kψ ∩ L∞ (Ω) as test function in (25), we can write Z Z gn (x, un , ∇un )(un − v) dx a(x, un , ∇un )∇(un − v) dx + Ω

Ω

(64)

≤ hf, un − vi.

This implies Z Z a(x, un , ∇un )∇(un − v0 ) dx + a(x, un , ∇un )∇(v0 − v) dx Ω Z Ω + gn (x, un , ∇un )(un − v) dx

(65)

Ω

≤ hf, un − vi.

By Fatou’s Lemma and the fact that a(x, un , ∇un ) * a(x, u, ∇u) weakly in

N Y

0

0

Lp (Ω, wi1−p ) one easily sees that

i=1

Z

Z a(x, u, ∇u)∇(u − v0 ) dx + a(x, u, ∇u)∇(v0 − v) dx Ω Z Ω + g(x, u, ∇u)(u − v) dx

(66)

Ω

≤ hf, u − vi.

Hence Z

Ω

a(x, u, ∇u)∇(u − v) dx +

Z

Ω

g(x, u, ∇u)(u − v) dx

≤ hf, u − vi.

(67)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

201

On the other hand, for h large enough we have Z Z N X ∂T (un ) p 0 k∞ |gn (x, un , ∇un )| dx ≤ b(h + kv0 k∞ )(c(x) + wi | h+kv∂x | ) dx i Ω Ω i=1 Z + |g(x, un , ∇un )| dx. {|un −v0 |>h}

(68) combining (61), (68) and the fact that un bounded in W01,p (Ω, w), we get Z |gn (x, un , ∇un )| dx ≤ c1 (69) Ω

taking also v0 as a test function in (25), we have Z gn (x, un , ∇un )un dx ≤ c2

(70)

Ω

by (69), (70) and Fatou’s Lemma we deduce g(x, u, ∇u) ∈ L1 (Ω), g(x, u, ∇u)u ∈ L1 (Ω). This proves Theorem 3.1. 5. Appendix Appendix 1: Proof of Lemma 4.2 From H¨ older’s inequality, the growth condition (20) we can show that A is bounded, and by using (26), we have Bn bounded. The coercivity follows from (7), (22) and (26). it remain to show that Bn is pseudo-monotone. Let a sequence (uk ) ∈ W01,p (Ω, w) such that uk * u weakly in W01,p (Ω, w), lim suphBn uk , uk − ui ≤ 0.

(71)

k→+∞

Let v ∈ W01,p (Ω, w), we will prove that lim inf hBn uk , uk − vi ≥ hBn u, u − vi. k→+∞

Since (uk )k is bounded in W01,p (Ω, w), by (20) we deduce that 0 QN 0 (a(x, uk , ∇uk ))k is bounded in i=1 Lp (Ω, wi1−p ), then there exists a funcQ 1−p0 p0 tion h ∈ N ) such that i=1 L (Ω, wi a(x, uk , ∇uk ) * h weakly in

N Y

i=1

0

0

Lp (Ω, wi1−p ),

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

202 0

0

similarly, it is easy to see that (gn (x, uk , ∇uk ))k is bounded in Lq (Ω, σ 1−q ), 0 0 then there exists a function ρn ∈ Lq (Ω, σ 1−q ) such that 0

0

gn (x, uk , ∇uk ) * ρn weakly in Lq (Ω, σ 1−q ). It is clear that, (15) Z Z h∇v dx a(x, uk , ∇uk )∇uk dx − lim inf hBn uk , uk − vi = lim inf k→+∞ k→+∞ Z Ω Ω + ρn (u − v) dx. Ω

(72)

On the other hand, by condition (21), we have Z (a(x, uk , ∇uk ) − a(x, uk , ∇u))(∇uk − ∇u) dx ≥ 0 Ω

which implies that Z Z Z a(x, uk , ∇uk )∇uk dx ≥ − a(x, uk , ∇u)∇u dx + a(x, uk , ∇uk )∇u dx Ω ZΩ Ω a(x, uk , ∇u)∇uk dx, + Ω

hence lim inf k→∞

Z

Ω

a(x, uk , ∇uk )∇uk dx ≥

Z

h∇u dx.

Combining (72) and (73), we get Z Z lim inf hBn uk , uk − vi ≥ h∇(u − v) dx + ρn (u − v) dx. k→+∞

Ω

Now, since v is arbitrary and lim

k→+∞

Z

(73)

Ω

(74)

Ω

Ω

gn (x, uk , ∇uk )(uk − u) dx = 0, we

have by using (71) and (74) Z a(x, uk , ∇uk )∇(uk − u) dx = 0 lim k→+∞

Ω

we deduce that Z lim (a(x, uk , ∇uk ) − a(x, uk , ∇u))∇(uk − u) dx = 0. k→+∞

Ω

In view of Lemma 4.1, we have ∇uk → ∇u a.e. in Ω, which with (74) yields lim inf hBn uk , uk − vi ≥ hBn u, u − vi. k→+∞

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

203

Thus complete the proof of Lemma 4.2. Appendix 2: We claim that vn ∈ Kψ (vn defined in (35)) 2 We have eγwn,k ≤ ck , let η = c1k Since vn = un − ηϕk (wn,k ), we remark that un if wn,k ≤ 0 vn ≥ un − wn,k if wn,k ≥ 0 then it suffices to prove that un − wn,k ≥ ψ we have  if |un | ≤ k  Tk (u) un − wn,k = un − k + Tk (u) if un ≥ k  un + k + Tk (u) if un ≤ −k which implies that

un − wn,k

  Tk (u) if |un | ≤ k ≥ Tk (u) if un ≥ k  un if un ≤ −k

since u ∈ Kψ , k ≥ kv0 k∞ then Tk (u) ≥ ψ, which implies that un −wn,k ≥ ψ. Finally since vn ∈ W01,p (Ω, w) then vn ∈ Kψ . References 1. L. Aharouch and Y. Akdim, Existence of solution of degenerated unilateral Problems with L1 data, Annale Math´ematique Blaise Pascal Vol. 11 (2004) pp 47-66. 2. L. Aharouch, Y. Akdim and E. Azroul, Quasilinear degenerate elliptic unilateral problems, AAA 2005:1 (2005) 11-31. DOI: 10.1155/AAA.2005.11 3. Y. Akdim, E. Azroul and A. Benkirane, Existence Results for Quasilinear Degenerated Equations via Strong Convergence of Truncations, Revista Matem´ atica Complutence, Madrid, (2004), 17; N 2, 359-379. 4. Y. Akdim, E. Azroul and A. Benkirane, Existence of solution for Quasilinear Degenerated Unilateral Problems, Annale Math´ematique Blaise Pascal Vol. 10 (2003) pp 1-20. 5. P. B´ enilan, L. Boccardo, T. Gallouet, R. Gariepy, M. Pierre and J. ´ zquez, An L1 -theory of existence and uniqueness of nonlinear elliptic L. Va equations, Ann. Scuola Norm. Sup. Pisa 22 (1995), 240-273. 6. L. Boccardo, F. Murat, and J. P. Puel , Existence of bounded solutions for nonlinear elliptic unilateral problems, Ann. Mat. Pura e Appl., 152 (1988) 183-196. 7. L. Boccardo, T. Gallou¨ et and F. Murat, A unified presentation of two existence results for problems with natural growth, in Progress in PDE, the Metz surveys 2, M. Chipot editor, Research in Mathematics, Longman, 296 (1993), 127-137.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

204

8. G. Dalmaso, F. Murat, L. Orsina and A. Prignet, Renormalized solutions of elliptic equations with general measure data, Ann. Scuola Norm. Sup Pisa Cl. Sci 12 4 (1999) 741-808. 9. P. Drabek, F. Nicolosi, Existence of bounded solutions for some degenerated quasilinear elliptic equations, Ann. Math. Pura Appl. (4) 165 (1993), 217238. 10. P. Drabek, A. Kufner and V. Mustonen, Pseudo-monotonicity and degenerated or singular elliptic operators, Bull. Austral. Math. Soc. 58 (1998), 213-221. 11. P. Drabek, A. Kufner and F. Nicolosi, Non linear elliptic equations, singular and degenerate cases, University of West Bohemia, (1996). 12. D. Gilbarg and N.S. Trudinger, Elliptic partial differential equations of second order, Springer-Verlag, Berlin, (1977). 13. A. Porretta, Existence for elliptic equations in L1 having lower order terms with natural growth,Portugal. Math. 57 (2000), 179-190. 14. A. Elmahi and D. Meskine , Unilateral elleptic problems in L1 with natural growth terms, Journal of Nonlinear and Convex Analysis, 5 N. 1, (2004), 97-112 15. J. Lions, Quelques m´ethodes de r´esolution des probl`emes aux limites non lin´eaires,Dunod, Paris (1969). 16. B. Opic and A. Kufner, Hardy-type inequalities, Pitman Research Notes in Mathematics Series vol. 219, Longman Scientific & Technical, Harlow, 1990, ISBN 0-582-05198-3.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

205

Existence and multiplicity results for some p(x)-Laplacian Neumann problems M. Bendahmane Al-Imam University, Faculty of Sciences, Department of Mathematics P. O. Box 90950, Riyadh 11623, Saudi Arabia E-mail: [email protected] M. Chrif Department of Mathematics, Faculty of Sciences, Atlas-Fes Fes, 3000, Morocco E-mail: [email protected] S. El Manouni Department of Mathematics, Faculty of Sciences, Al-Imam University P. O. Box 90950, Riyadh, 11623, Saudi Arabia E-mail: [email protected] In this paper, using an equivalent variational approach to a recent Ricceri’s three critical points theorem,13 we obtain the existence of at least three nontrivial solutions of a Neumann problem for elliptic equations with variable exponents. Keywords: p(x)-Laplacian; Elliptic equations; Neumann conditions; Three critical points theorem.

1. Introduction Let Ω be a bounded domain of IRN , N ≥ 2 with a smooth boundary ∂Ω. We consider the following Neumann problem for the corresponding elliptic problem −div(|∇u|p(x)−2 ∇u) + |u|p(x)−2 u = λf (x, u) + µg(x, u) in Ω (P1 ) ∂u on ∂Ω ∂ν = 0 where p ∈ C(Ω), p(x) > 1 for every x ∈ Ω. f is a C 1 -function on Ω × [0, ∞) satisfying (1.1)

|f (x, t)| ≤ c1 tα(x)−1 + c2

∀(x, t)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

206

for some α ∈ C+ (Ω) with α + < p− ,

(1.2) where

h+ = sup h(x) and h− = inf h(x) ∀h ∈ C+ (Ω). x∈Ω

x∈Ω

(1.3)

f (x, t) < 0, for all t ∈]0, 1[,

(1.4)

f (x, t) > M, for all t > t0

where M is a positive constant and t0 > 1. The function g is assumed to be a measurable function with respect to x in Ω for every t in IR, and is a C 1 -function with respect to t in IR for almost every x in Ω and satisfies (1.5)

sup |g(x, t)| ∈ L1 (Ω)

|t|≤s

for all s > 0. Here Ω is a bounded domain in IRN with smooth boundary ∂Ω, div(|∇u|p(x)−2 ∇u) is the p(x)−Laplacian, the generalization of the classical p−Laplacian operator, and ν is the outward unit normal to ∂Ω. Recently, elliptic equations with variable exponents have been extensively investigated and have received much attention. They have been the subject of recent developments in nonlinear elasticity theory and electrorheological fluids dynamics Ru˜zicka.15 In that context, let us mention that there appeared a series of papers on problems which lead to spaces with variable exponent. We refer the reader to Fan et al.,8,9 Ru˜zicka16 and the references therein. Let us point out that when p(x) = p = constant, there is a large literature which deal with problems involving the p-Laplacian with Dirichlet boundary conditions for elliptic equations in bounded or unbounded domains and we do not need here to cite them since the reader may easily reach such papers. Note also that many papers deal with problems related to the pLaplacian with Neumann conditions. We can cite, among others, the articles Anello et al.1 and Bonanno et al.4 and the references therein for details. The case of p(x)−Laplacian with Neumann conditions has been studied by Dai,5 Mihˇ ailescu10 and Shi and Ding,17 all the authors used a past result of Ricceri.12 Finally, it would be interesting at this stage to refer the reader to the recent work Manouni et al.6 where the authors established the existence

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

207

of at least three nontrivial solutions for elliptic systems involving the pLaplacian with Neumann boundary conditions. Our objective is to study the Neumann problem for such an equation of the type (P1 ). Precisely, based on a recent result due to Ricceri,13 we are interested in the existence and multiplicity of weak nontrivial solutions for the problem (P1 ) in the Sobolev space W 1,p(x) (Ω). Along this paper we fix p− > N. Recall that a weak solution of system (P1 ) is any u ∈ W 1,p(x) (Ω) such that Z Z p(x)−2 p(x)−2 (|∇u| ∇u∇ϕ + |u| uϕ) dx = (λf (x, u) + µg(x, u))ϕ dx Ω

Ω

for all ϕ ∈ W

1,p(x)

(Ω).

Remark 1.1. Let us remark that (1.2) and (1.5) guarantees that integrals given in the right side are well defined. In particular, we will consider the following Neumann problem −∆p(x) u + |u|p(x)−2 u = λ(|u|α(x)−2 u − 1) + µ|u|γ(x)−2 u 0 (P1 ) ∂u on ∂Ω, ∂ν = 0

in Ω

where γ ∈ C+ (Ω), γ > 1

(1.6) and α satisfies

α + < p− .

(1.7)

To prove the existence of at least three weak solutions for each of the given problems (P1 )0 and (P1 ), we will use the following result proved in Ricceri13 that, on the basis of Bonanno,2 can be equivalently stated as follows Theorem 1.1. Let X be a reflexive real Banach space; Φ : X −→ IR is bounded on each bounded subset of X, continuously Gˆ ateaux differentiable and sequentially weakly lower semi-continuous functional whose Gˆ ateaux ∗ derivative admits a continuous inverse on X ; J : X −→ IR a continuously Gˆ ateaux differentiable functional whose Gˆ ateaux derivative is compact. Moreover, assume that (i)

lim

kuk−→∞

(Φ(u) + λJ(u)) = +∞ for all λ ∈ ]0, +∞[

and that there are r ∈ IR, u0 , u1 ∈ X such that (ii) Φ(u0 ) < r < Φ(u1 )

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

208

(iii)

inf

u∈Φ−1 (]−∞,r])

J(u) >

(Φ(u1 ) − r)J(u0 ) + (r − Φ(u0 ))J(u1 ) . Φ(u1 ) − Φ(u0 )

Then, there exist an open interval A of ]0, +∞[ and a positive real number t such that for every λ ∈ A, and every continuously Gˆ ateaux differentiable functional Ψ : X −→ IR with compact derivative, there exists δ > 0 such that for each µ ∈ [0, δ], the equation 0

0

0

Φ (u) + λJ (u) + µΨ (u) = 0 has at least three solutions in X whose norms are less than t. 2. Preliminaries We list some well known definitions and basic properties and recall some background facts concerning generalized Lebesgue-Sobolev spaces Lp(x) (Ω), 1,p(x) W 1,p(x) (Ω) and W0 (Ω), where Ω is an open subset of IRN and introduce some notations used below. For more details about these spaces, we refer the reader to the book of Musielak11 and the papers of Kov´ acˇik et al.7 and 8,9 Fan et al. Set ∞ L∞ + (Ω) = {h; h ∈ L (Ω), ess inf h(x) > 1 for all x ∈ Ω}. x∈Ω

For any h ∈

L∞ + (Ω)

we define

h+ = ess sup h(x) x∈Ω

and h− = ess inf h(x). x∈Ω

For any p(x) ∈ L∞ + (Ω), we define the variable exponent Lebesgue space Lp(x)R(Ω) = {u : uis a measurable real-valued function such that |u(x)|p(x) dx < ∞}. Ω

We define a norm, the so-called Luxemburg norm, on this space by the formula Z u(x) p(x) dx ≤ 1 . |u|p(x) = inf µ > 0; µ Ω

Variable exponent Lebesgue spaces resemble classical Lebesgue spaces in many respects: they are Banach spaces (Theorem 2.5, Kov´ acˇik7 ), the H¨ older 7 inequality holds (Theorem 2.1, Kov´ acˇik ), they are reflexive if and only if 1 < p− ≤ p+ < ∞ (Corollary 2.7, Kov´ acˇik7 ) and continuous functions + are dense if p < ∞ (Theorem 2.11, Kov´ acˇik7 ). The inclusion between

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

209

Lebesgue spaces also generalizes naturally (Theorem 2.8, Kov´ acˇik 7 ): if 0 < |Ω| < ∞ and r1 , r2 are variable exponents so that r1 (x) ≤ r2 (x) almost everywhere in Ω then there exists the continuous embedding Lr2 (x) (Ω) ,→ 0 Lr1 (x) (Ω), whose norm does not exceed |Ω| + 1. We denote by Lp (x) (Ω) the conjugate space of Lp(x) (Ω), where 1/p(x) + 1/p0 (x) = 1. For any u ∈ 0 Lp(x) (Ω) and v ∈ Lp (x) (Ω) the H¨ older type inequality Z 1 1 uv dx ≤ − + 0 − |u|p(x) |v|p0 (x) p p Ω holds true. An important role in manipulating the generalized LebesgueSobolev spaces is played by the modular of the Lp(x) (Ω) space, which is the mapping ρp(x) : Lp(x) (Ω) → IR defined by ρp(x) (u) =

Z

Ω

|u|p(x) dx.

If u ∈ Lp(x) (Ω) and p+ < ∞ then the following relations hold −

+

+

−

|u|p(x) > 1 ⇒ |u|pp(x) ≤ ρp(x) (u) ≤ |u|pp(x) ; |u|p(x) < 1 ⇒ |u|pp(x) ≤ ρp(x) (u) ≤ |u|pp(x) ; |un − u|p(x) → 0 ⇔ ρp(x) (un − u) → 0. We also consider the weighted variable exponent Lebesgue spaces. We define also the variable Sobolev space W 1,p(x) (Ω) = {u ∈ Lp(x) (Ω) : |∇u| ∈ Lp(x) (Ω)}. On W 1,p(x) (Ω) we may consider one of the following equivalent norms kukp(x) = |u|p(x) + |∇u|p(x) or

Set

kuk = inf µ > 0;

Z

Ω

Ip(x) (u) =

Z

∇u(x) p(x) u(x) p(x) + dx ≤ 1 . µ µ Ω

|∇u|p(x) + |u|p(x) dx.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

210

For all u ∈ W 1,p(x) (Ω) the following relations hold −

+

+

−

(2.1)

kuk > 1 ⇒ kukp ≤ Ip(x) (u) ≤ kukp ;

(2.2)

kuk < 1 ⇒ kukp ≤ Ip(x) (u) ≤ kukp .

Finally, we remember some embedding results regarding variable exponent Lebesgue-Sobolev spaces. For the continuous embedding between variable exponent Lebesgue-Sobolev spaces we refer to (Theorem 1.1, Fan et al.8 ): if p : Ω → IR is Lipschitz continuous and p+ < N , then for any q ∈ L∞ + (Ω) N p(x) 1,p(x) with p(x) ≤ q(x) ≤ N −p(x) , there is a continuous embedding W (Ω) ,→

Lq(x) (Ω). In what concerns if N < p− = ess inf x∈Ω p(x) ≤ p(x) for any x ∈ Ω, by Theorem 2.2 in Fan et al.9 we deduce that W 1,p(x) (Ω) is contin− − uously embedded in W 1,p (Ω). Since N < p− it follows that W 1,p (Ω) is compactly embedded in C(Ω). Thus, we obtain that W 1,p(x) (Ω) is continuously embedded in C(Ω).

From now on, E is the space W 1,p(x) (Ω) with the norm Z 1/p(x) kuk = , |∇u|p(x) + |u|p(x) dx Ω

which is clearly equivalent to the usual one, on the space C(Ω) we consider the norm kuk∞ = supx∈Ω¯ |u(x)|. When p− > N, Sobolev’s Theorem implies that E is compactly embedded in C(Ω), hence (2.3)

c=

kuk∞ < +∞. u∈E\{0} kuk sup

Proposition 2.1. Let I : E → E ∗ be the operator defined by Z |∇u|p(x)−2 ∇u∇ϕ + |u|p(x)−2 uϕ dx I(u)ϕ = Ω

for all u, ϕ ∈ E. Then I admits a continuous inverse on E ∗ .

Proof. Denoting by h., .i the usual inner product in IRN . We have for some positive constant C1

Hence,

Ip(x) (u) − hI(u), ui ≥ C1 ≥ C1 kukp −1 . kuk kuk lim

kuk→+∞

hI(u), ui = +∞. kuk

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

211

Then I is coercive. Moreover, by simple arguments we can easily verify that I is hemicontinuous. It remains to show that I is uniformly monotone. Indeed, recall first the well known inequality |x − y|m ≤ 2m h|x|m−2 x − |y|m−2 y, x − yi, ∀x, y ∈ IRN , ∀m ≥ 2. Thus, it is easy to see that hI(u1 ) − I(u2 ), u1 − u2 i ≥

1 2 p+

Z

Ω

|∇u1 − ∇u2 |p(x) + |u1 − u2 |p(x) dx +

≥ c(p+ )ku1 − u2 kp , for every u1 and u2 belonging to E. This means that I is uniformly monotone operator in E. Therefore, the conclusion of Proposition 2.1 follows directly from the result (Theorem 26. A) of Zeidler.19 Our main result are the following Theorem 2.1. Suppose p− > N and let α satisfying (1.2). Assume that f : Ω × IR → IR is a function which is measurable in Ω and C 1 in IR satisfying (1.1), (1.3) and (1.4). Then, there exist an open interval A of ]0, +∞[ and a positive real number t such that, for every λ ∈ A and every function g : Ω × IR → IR which is measurable in Ω and C 1 in IR satisfying (1.5), there exists σ > 0 such that for each µ ∈ [0, σ] the system (P1 ) has at least three weak solutions whose norms in E are less than t. 3. Proof of the main result We begin by setting Φ(u) =

Z

Ω

1 (|∇u|p(x) + |u|p(x) ) dx p(x)

J(u) = − and Ψ(u) =

Z

Ω

Z

F (x, u) dx Ω

1 |u|γ(x) dx γ(x)

for each u ∈ E. It is well known that Φ and J are well defined and continuously Gˆ ateaux differentiable with Z 0 Φ (u)ϕ = (|∇u|p(x)−2 ∇u∇ϕ + |u|p(x)−2 uϕ) dx Ω

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

212

and 0

J (u)ϕ = − 0

Z

f (x, u)ϕ dx

Ω

for all u, ϕ in E. Note that J is compact and Φ is clearly weakly lower semi-continuous and bounded on each bounded subset of E. Proposition 0 2.1 ensures that Φ admits a continuous inverse on E ∗ . Moreover, it is easy to see that lim

kuk−→+∞

(Φ(u) + λJ(u)) = +∞

for all λ ∈ ]0, +∞[. Indeed, since E is embedded in C 0 (Ω), we have + c1 J(u) ≥ −|Ω|( − kukα ∞ + c2 kuk∞ ), α where |Ω| denotes the measure of Ω. Then, the coercivity of Φ + λJ can be easily deduced from (2.1) since α+ < p− and λ > 0 and from the fact that kuk∞ ≤ Dkuk for some positive constant D. Then (i) is verified. In the sequel, we will verify the conditions (ii) and (iii) of Theorem 1.1. By (1.3) and (1.4), it follows that F (x, t) is increasing for t ∈ (1, ∞) and decreasing for t ∈ (0, 1), uniformly with respect to x and F (x, 0) = 0 is obvious. Also, by (1.4), we can choose δ > 1 such that F (x, u) ≥ 0 = F (x, 0) ≥ F (x, τ ), ∀u > δ, τ ∈ ]0, 1[. Let b, d be two numbers such that 0 δ such that dp |Ω| > 1. Then we obtain Z + Z 1 bp F (x, d) dx. sup F (x, u) ≤ 0 < p+ p− c d Ω Ω 0≤u≤b Define the real number r by r=

1 p+

By choosing

p+ b . c

u0 (x) = 0 and u1 (x) = d for every we have Φ(u0 ) = J(u0 ) = 0, Φ(u1 ) = and J(u1 ) = −

Z

Ω

Z

Ω

x∈Ω

1 |d|p(x) dx p(x)

F (x, d) dx.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

213

Then we clearly have Φ(u1 ) ≥

1 p− d |Ω| > r. p+

Thus we deduce that Φ(u0 ) < r < Φ(u1 ). Then (ii) in Theorem 1.1 is verified. On the other hand, we have (Φ(u1 ) − r)J(u0 ) + (r − Φ(u0 ))J(u1 ) − = rZ Φ(u1 ) − Φ(u0 )

Z Ω

F (x, d) dx Ω

1 |d|p(x) dx p(x)

> 0.

We also have 1 Ip(x) (u) ≤ Φ(u), p+ which implies that Ip(x) (u) ≤ p+ r < 1, for all x ∈ Ω and for all u ∈ E such that Φ(u) ≤ r. Using (2.1), we get kuk ≤ 1. This implies that 1 kuk ≤ Φ(u) ≤ r. p+ Taking into account that 1

|u(x)| ≤ c(p+ r) p+ < b ∞ for all x ∈ Ω and for all u ∈ E such that Φ(u) ≤ r, with c = supu∈E kuk kuk . It follows Z − inf J(u) = sup −J(u) ≤ sup F (x, u) dx ≤ 0. −1

u∈Φ

(]−∞,r])

Φ(u)≤r

Ω 0≤u≤b

Consequently, we obtain inf

u∈Φ−1 (]−∞,r])

J(u) >

(Φ(u1 ) − r)J(u0 ) + (r − Φ(u0 ))J(u1 ) . Φ(u1 ) − Φ(u0 )

This means that condition (iii) in Theorem 1.1 is verified. Moreover, since the function G : Ω × IR −→ IR is a measurable in Ω and C 1 in IR × IR satisfying the condition (1.5), then the functional

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

214

Ψ(u) =

Z

Ω

Z

u(x)

g(x, t) dt 0

!

dx is well defined and continuously Gˆ ateaux

differentiable on E, with compact derivative, and one has Z 0 g(x, u(x))ϕ(x) dx Ψ (u)ϕ = Ω

for all u, ϕ ∈ E. So, in view of Proposition 2.1 and Theorem 1.1, the proof of Theorem 2.1 is achieved. Corollary 3.1. Suppose p− > N. Let α satisfying (1.2). Then there exists an open interval A of ]0, +∞[ and a positive real number t such that, for every λ ∈ A and every γ, with γ > 1 there exists σ > 0 such that for each µ ∈ [0, σ] the system (P1 )0 has at least two weak nonzero solutions whose norms in E are less than t. Proof. Let Φ(u) =

Z

Ω

1 (|∇u|p(x) + |u|p(x) ) dx p(x) Z

(

Ψ(u) =

Z

J(u) = −

Ω

and

1 |u|α(x) − u) dx α(x)

Ω

1 |u|γ(x) dx γ(x)

for each u ∈ E. It is well known that Φ and J are well defined and continuously Gˆ ateaux differentiable with Z 0 Φ (u)ϕ = (|∇u|p(x)−2 ∇u∇ϕ + |u|p(x)−2 uϕ) dx Ω

and 0

J (u)ϕ = −

Z

Ω

|u|α(x)−2 uϕ dx

0

for all u, ϕ in E. Note that J is compact and Φ is clearly weakly lower semi-continuous and bounded on each bounded subset of E. Proposition 0 2.1 ensures that Φ admits a continuous inverse on E ∗ . Moreover, it is easy to see that lim

kuk−→+∞

(Φ(u) + λJ(u)) = +∞

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

215

for all λ ∈ ]0, +∞[. Consider now H(x, u) =

1 |u|α(x) − u. α(x)

Since α(x) > 0 for all x ∈ Ω we have lim H(x, u) = +∞.

|u|→∞

Choose δ > 1 such that 1 |u|α(x) − u > 0 α(x) for all u > δ. Hence we get H(x, u) ≥ 0 = H(x, 0) ≥ H(x, τ ), ∀u > δ, τ ∈ ]0, 1[. Let us choose b, d two real numbers and u0 and u1 the same as in the proof of Theorem 2.1. Then we obtain (ii) and (iii) of Theorem 1.1 when adapting the techniques given in the proof of Theorem 2.1. On the other hand, the functional Z 1 |u|γ(x) dx Ψ(u) = γ(x) Ω

is continuously Gˆ ateaux differentiable on E, with compact derivative. So, in view of Theorem 1.1, the proof is completed.

Acknowledgement The research of S. El Manouni is supported by Al-Imam University project No. 28/12. References 1. G. Anello and G. Cordaro, Existence of solutions of the Neumann problem involving the p-Laplacian via variational principle of Ricceri. Arch. Math. (Basel)79 (2002)274-287. 2. G. Bonanno, A minimax inequality and its applications to ordinary differential equations. J. Math. Anal. Appli. 270(2002) 210-219. 3. G. Bonanno, Some remarks on a three critical points theorem, Nonlinear Anal. 54(2003) 651-665. 4. G. Bonanno and P. Candito, Three solutions to a Neumann problem for elliptic equations involving the p-Laplacian. Arch. Math. 80(2003) 424-429. 5. G. Dai, Three solutions for a Neumann-type differential inclusion problem involving the p(x)-Laplacian, Nonlinear Anal. to appear 6. S. El Manouni and M. Kbiri Alaoui, A result on elliptic systems with Neumann conditions via Ricceri’s three critical points theorem, Nonlinear Anal. to appear

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

216

7. O. Kov´ acˇik and J. R´ akosn´ık, On spaces Lp(x) and W 1,p(x) , Czechoslovak Math. J. 41(1991) 592-618. 8. X. L. Fan, J. Shen and D. Zhao, Sobolev embedding theorems for spaces W k,p(x) (Ω), J. Math. Anal. Appl. 262(2001) 749-760. 9. X. L. Fan and D. Zhao, On the spaces Lp(x) (Ω) and W m,p(x) (Ω), J. Math. Anal. Appl. 263(2001) 424-446. 10. M. Mihˇ ailescu, Existence and multiplicity of solutions for a Neumann problem involving the p(x)-Laplace operator, Nonlinear Anal., 67(2007) 14191425. 11. J. Musielak, Orlicz spaces and modular spaces. Lecture Notes in Mathematics, 1034. Springer-Verlag, Berlin (1983). 12. B. Ricceri, On three critical points theorem, Arch. Math. (Basel) 75 (2000), 220-226. 13. B. Ricceri, A three critical points theorem revisited. Nonlinear Anal. to appear (2008). 14. B. Ricceri, A general variational principle and some of its applications. J. Comput. Appl. Math. 113(2000) 401-410. 15. M. Ruzicka, Electrorheological fluids: Modelling and Mathematical Theory, Lecture Notes in Math., Springer-Verlag, Berlin (2002). 16. M. Ruzicka, Flow of shear dependent electrorheological fluids. C. R. A. S. Sci. Paris. S´er. I. Math. 329(1999) 393-398. 17. X. Shi and X. Ding, Existence and multiplicity of solutions for a general p(x)-Laplacian Neumann problem, Nonlinear Anal. to appear 18. J. Simon, r´egularit´e de la solution d’une equation non lin´eaire dans IR N . LMN 665, P. Benilan ed., Berlin-Heidelberg-New York 1978. 19. E. Zeidler, Nonlinear functional analysis and its applications. Vol. II/B. Berlin-Heidelberg-New York 1978.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

217

Positive solutions with changing sign energy to nonhomogeneous elliptic problem of fourth order M. Talbi∗ and N. Tsouli† D´ epartement de Math´ ematiques et Informatique Facult´ e des Sciences Universit´ e Mohamed 1, Oujda, Morocco E-mails: ∗ talbi [email protected], † [email protected] In this paper, we study the existence for two positive solutions to a nonhomogeˆ neous elliptic equation of fourth order with a parameter λ such that 0 < λ < λ. The first solution has a negative energy while the energy of the second one is ˆ The values λ0 and λ ˆ positive for 0 < λ < λ0 and negative for λ0 < λ < λ. are given under variational form and we show that every corresponding critical point is solution of the nonlinear elliptic problem (with a suitable multiplicative term). Keywords: Ekeland’s principle; p-Laplacian operator; Palais-Smale condition.

1. Introduction We consider the problem with Navier boundary conditions 2 ∆p u = λ|u|q−2 u + |u|r−2 u in Ω (Pλ ) u = ∆u = 0 on ∂Ω. Here Ω is a smooth domain in RN (N ≥ 1), ∆2p is the p-biharmonic operator defined by ∆2p u = ∆(| ∆u |p−2 ∆u), λ is a positive parameter, p, q and r are reals such that   p∗ = N p if p < N/2, ∗ 2 1 < q < p < r < p2 , where N − 2p  p∗ = +∞ if p ≥ N/2. 2

Such kind of problems with combined concave and convex nonlinearities were studied recently by several authors2–7,9–11,17 in the case of operator ∆p . Our main results here can be summarized as follows: Let us put X = W02,p (Ω) ∩ W 2,p (Ω). We find two characteristic values λ0

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

218

ˆ (λ0 < λ) ˆ under variational form, i.e. and λ (V )

λ0 = C0 (p, q, r)

inf

u∈X\{0}

F (u) and

ˆ = C(p, ˆ q, r) λ

inf

u∈X\{0}

F (u),

ˆ (the such that two branches of positive solutions to (Pλ ) exist for λ ∈ ]0, λ[ functional F will be given below). Moreover, the energy of the first positive ˆ while the energy of the second positive solution is negative for λ ∈]0, λ[ solution changes sign at λ0 , i.e. it is positive forλ ∈ ]0, λ0 [ and negative for ˆ Notice that these two positive solutions are found simultaneously λ ∈ ]λ0 , λ[. and that our approach does not use the mountain-pass lemma. On the other hand, we show that every solution of (V ) is a solution of the problem (Pλ ) (with a suitable multiplicative term). This second point lets expect that the first nonlinear eigenvalue ζ of (V ), i.e. ζ = sup{λ > 0 : (Pλ )

has a nonnegative solution}

may satisfy a variational problem similar to (V ) (see4 for p = 2). Let us ˆ coincides with ζ when q → p and that λ ˆ constitutes a good precise that λ minoration of ζ in the general case 1 < q < p. We consider the transformation of Poisson problem used by P.Dr´ abek and M.ˆ otani (cf.12 ): We recall some properties of the Dirichlet problem for the Poisson equation: −∆u = f in Ω, (1) u = 0 on ∂Ω. It is well known that (1) is uniquely solvable in W 2,p (Ω) ∩ W01,p (Ω) for all f ∈ Lp (Ω) and for any p ∈ ]1, +∞[. We denote by : 1,p X = W 2,p Z (Ω) ∩ W0 (Ω),

kukp = (

Ω

|u|p dx)1/p the norm in Lp (Ω),

kuk2,p = (k∆ukpp + kukpp )1/p the norm in X, kuk∞ the norm in L∞ (Ω), 0 and < ., . > is the duality bracket between Lp (Ω) and Lp (Ω), where p0 = p/(p − 1). Denote by Λ the inverse operator of −∆ : X → Lp (Ω). The following lemma gives us some properties of the operator Λ (cf.,1216 ) Lemma 1.1. (i) (Continuity): There exists a constant cp > 0 such that kΛf k2,p ≤ cp kf kp

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

219

holds for all p ∈ ]1, +∞[ and f ∈ Lp (Ω). (ii) (Continuity) Given k ∈ N∗ , there exists a constant cp,k > 0 such that kΛf kW k+2,p ≤ cp,k kf kW k,p holds for all p ∈ ]1, +∞[ and f ∈ W k,p (Ω). (iii) (Symmetry) The following identity: Z Z u · Λvdx Λu · vdx = Ω

Ω

0

holds for all u ∈ Lp (Ω) and v ∈ Lp (Ω) with p ∈ ]1, +∞[. ¯ for all α ∈ ]0, 1[; (iv) (Regularity) Given f ∈ L∞ (Ω), we have Λf ∈ C 1,α (Ω) moreover, there exists cα > 0 such that kΛf kC 1,α ≤ cα kf k∞ . ¯ and f ≥ 0 (v) (Regularity and Hopf-type maximum principle) Let f ∈ C(Ω) 1,α ¯ then w = Λf ∈ C (Ω), for all α ∈ ]0, 1[ and w satisfies: w > 0 in Ω, ∂w ∂n < 0 on ∂Ω. (vi) (Order preserving property) Given f, g ∈ Lp (Ω) if f ≤ g in Ω, then Λf < Λg in Ω. Remark 1.1. (∀u ∈ X)(∀v ∈ Lp (Ω))

v = −∆u ⇐⇒ u = Λv.

Let us denote Np the Nemytskii operator defined by Np (v)(x) = |v(x)|p−2 v(x) if v(x) 6= 0 Np (v)(x) = 0 if v(x) = 0, 0

and we have ∀v ∈ Lp (Ω), ∀w ∈ Lp (Ω) : Np (v) = w ⇐⇒ v = Np0 (w). We define the functionals P, Q, R : Lp (Ω) → R as follows: P (v) = kvkpp ,

Q(v) = ||Λv||qq

and R(v) = ||Λv||rr .

The operator Λ enables us to transform problem (Pλ ) to an other problem which we will study in the space Lp (Ω). Definition 1.1. We say that u ∈ X \ {o} is a solution of problem (Pλ ), if v = −∆u is a solution of the following problem Find v ∈ Lp (Ω) \ {o}, such that 0 (Pλ ) 0 Np (v) = λΛ(Nq (Λv)) + Λ(Nr (Λv)) in Lp (Ω).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

220

For solutions of (Pλ ) we understand critical points of the associated EulerLagrange functional Eλ ∈ C 1 (Lp (Ω)), given by Eλ (v) =

1 1 1 P (v) − λ Q(v) − R(v). p q r

As in (cf.13,19 ), we introduce the modified Euler-Lagrange functional defined ˜λ (t, v) = E(tv). If v is an arbitrary element of Lp (Ω), on R × Lp (Ω) by E ∂t E˜λ (., v) (resp. ∂tt E˜λ (., v))are the first (resp. second) derivative of the real ˜λ (t, v). valued function: t 7→ E 2. Preliminary results ˜λ is even in t and that we are interested by the Since the functional E positive solutions, we limit our study for t > 0. Lemma 2.1. For every v ∈ Lp (Ω) \ {0}, There is a unique λ(v) > 0 such ˜λ (t, v) has exactly two positive zeros that the real valued function t 7→ ∂ E (resp. one positive zero ) if 0 < λ < λ(v) (resp. λ = λ(v)). This function has no zero for λ > λ(v). Proof : Let v be an arbitrary element of Lp (Ω) \ {0} and let us write ∂t E˜λ (t, v) = tq−1 F˜λ (t, v),

where F˜λ (t, v) = tp−q P (v) − λQ(v) − tr−q R(v).

Then ∂tt E˜λ (t, v) = (q − 1)tq−2 F˜λ (t, v) + tq−1 ∂t F˜λ (t, v), holds true, with ∂t F˜λ (t, v) = tp−q−1 [(p − q)P (v) − (r − q)tr−p R(v)]. It is clair that the real valued function t 7→ F˜λ (t, v) is increasing on ]0, t(v)[, decreasing on ]t(v), +∞[ and attains its unique maximum for t = t(v), where t(v) = (

1 p − q P (v) r−p . ) r − q R(v)

(2)

Thus, if F˜λ (t(v), v) > 0 (resp. F˜λ (t(v), v) = 0), the function t 7→ F˜λ (t, v) has two positive zeros (resp. one positive zero) and has no zero if F˜λ (t(v), v) < 0. On the other hand, a direct computation gives r−q r − p p − q P (v) r−p F˜λ (t(v), v) = ( ) R(v) − λQ(v). p − q r − q R(v)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

221

We deduce that F˜λ (t(v), v) > 0 (resp. F˜λ (t(v), v) < 0) for λ < λ(v) (resp. λ > λ(v)) and F˜λ(v) (t(v), v) = 0, where r−q

λ(v) = cˆ

P r−p (v) p−q

Q(v)R r−p (v)

,

(3)

with cˆ =

r−q r − p p − q r−p . ( ) p−q r−q

˜λ (t, v) has two Hence, if λ ∈ ]0, λ(v)[, the real valued function t 7→ ∂t E positive zeros, denoted by t1 (v, λ) and t2 (v, λ), verifying 0 < t1 (v, λ) < t(v) < t2 (v, λ). Since F˜λ (t1 (v, λ), v) = F˜λ (t2 (v, λ), v) = 0, ∂t F˜λ (t, v) > 0 for t < t(v) and ∂t F˜λ (t, v) < 0 for t > t(v), it follows that ˜λ (t2 (v, λ), v) < 0. ∂tt E˜λ (t1 (v, λ), v) > 0 and ∂tt E

(4)

˜λ (t, v), (t > 0) achieves This means that the real valued function t 7→ E its unique local minimum at t = t1 (v, λ) and its global maximum at t = t2 (v, λ). ˆ= Lemma 2.2. If we put λ

inf

v∈Lp (Ω)\{0}

ˆ > 0. λ(v), then λ

Proof : By Sobolev injection theorem, we have X ,→ Lq (Ω) and X ,→ Lr (Ω). Thus there exists two positive constants c1 and c2 such that ||Λv||q ≤ c1 ||v||p

et ||Λv||r ≤ c2 ||v||p .

Then (3) implies for every v ∈ Lp (Ω) \ {0} λ(v) ≥

cˆ r(p−q) r−p

> 0.

cq1 c2

ˆ and let (vn ) be minimizing sequence of v Consider λ ∈ ]0, λ[ ˜ Eλ (t1 (v, λ), v) in Lp (Ω) \ {0} (resp. of v 7→ E˜λ (t2 (v, λ), v)). Put Vn = t1 (vn , λ)vn and Wn = t2 (vn , λ)vn . Lemma 2.3. The sequences (Vn ) and (Wn ) verify : (i)

lim sup ||Vn ||p < +∞ n→+∞

(ii)

lim inf ||Vn ||p > 0 n→+∞

(resp. lim sup ||Wn ||p < +∞ ) n→+∞

(resp. lim inf ||Wn ||p > 0) n→+∞

7→

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

222

˜λ [t1 (vn , λ), vn ) = 0. Proof : (i) We know that ∂t E Hence ||Vn ||pp = λ||ΛVn ||qq + ||ΛVn ||rr .

(5)

Suppose that there is a subsequence of (Vn ), still denoted by (Vn ) such that lim ||Vn ||p = +∞. Us r > q, there exist a constant c > 0 such that

n→+∞

||ΛVn ||q ≤ c||ΛVn ||r . Then the relation (5) implies that +∞. The fact that 0 < q < r enables us to deduce: Then

lim ||ΛVn ||r = n→+∞ q ||ΛVn ||q = on (||ΛVn ||rr ).

||Vn ||pp = ||ΛVn ||rr (1 + on (1)), and 1 1 Eλ (Vn ) = ||ΛVn ||rr ( − + on (1)). p r which implies that Eλ (Vn ) tends to +∞ as n goes to +∞ and this is impossible. The same arguments with a minimizing sequence (vn ) of v 7→ E˜λ (t2 (v, λ), v) show that lim sup ||Wn ||p < +∞. n→+∞

(ii) Relation (5) and the fact that ∂tt E˜λ [t1 (vn , λ), vn ) > 0, implies (p − 1)||Vn ||pp − λ(q − 1)||ΛVn ||qq − (r − 1)||ΛVn ||rr > 0.

(6)

If we combine (5) and (6), we obtain for every n ∈ N λ(p − q)||ΛVn ||qq + (p − r)||ΛVn ||rr > 0. So q−p r−p Q(Vn ) + R(Vn ) pq pr −1 (λ(p − q)Q(Vn ) + (p − r)R(vn )) ≤ pq < 0.

Eλ (Vn ) = λ

Suppose that there is a subsequence of (Vn ), still denoted by (Vn ) such that lim ||Vn ||p = 0. By Sobolev injection theorem we den→+∞

duce that

lim ||ΛVn ||q = 0 and

n→+∞

lim Eλ (Vn ) = 0, i.e

n→+∞

inf

v∈Lp (Ω)\{0}

lim ||ΛVn ||r = 0. It follows that

n→+∞

˜λ (t1 (v, λ), v) = 0, which is impossible E

˜λ (t1 (vn , λ), vn ) < 0 for every n. since E

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

223

Let (vn ) be a minimizing sequence of v 7→ E˜λ (t2 (v), v) in Lp (Ω) \ {0}. ˜λ (t2 (vn ), vn ) = 0 and ∂tt E˜λ (t2 (vn ), vn ) < 0, it follows that Sinse ∂t E ||Wn ||pp −λ||ΛWn ||qq − ||ΛWn ||rr = 0, (p − 1) ||Wn ||pp − λ(q − 1)||ΛWn ||qq − (r − 1)||ΛWn ||rr < 0. Combining the two last inequalities and by Sobolev injection theorem there exist a constant c0 such that for every n we have (p − q)||Wn ||pp < (r − q)||ΛWn ||rr ≤ c0 ||Wn ||rp . Hence (p − q) ≤ c0 ||Wn ||pr−p . Now, suppose that there is a subsequence of (Wn ), still denoted by (Wn ) such that lim ||Wn ||p = 0. This implies that p−q ≤ 0. which is impossible n→+∞

since p > q.

˜λ (t1 (v, λ), v) and v 7→ E ˜λ (t2 (v, λ), v) Lemma 2.4. The functionals v 7→ E p are bonded below in L (Ω). Proof : Let (vn ) be a minimizing sequence of the functional v 7→ ˜λ (t1 (v, λ), v). E ˜λ (t1 (vn , λ), vn ) = 0, then We know that ∂t E [t1 (vn , λ)]p ||vn ||pp = λ[t1 (vn , λ)]q ||Λvn ||qq + [t1 (vn , λ)]r ||Λvn ||rr . Hence ˜λ (t1 (vn , λ), vn ) = λ( 1 − 1 )[t1 (vn , λ)]q ||Λvn ||qq +( 1 − 1 )[t1 (vn , λ)]r ||Λvn ||rr . E p q p r As p < r, we conclude that 1 1 E˜λ (t1 (vn , λ), vn ) ≥ λ( − )[t1 (vn , λ)]q ||Λvn ||qq . p q

(7)

Sobolev injection of X in Lq (Ω) and the fact that lim sup ||Vn ||p < +∞, n→+∞

implies that there exists c and k positive such that for every n in N, we have ||Vn ||p < k. and ||ΛVn ||q ≤ c||Vn ||p < kc. As q < p, the inequality (7) implies 1 1 E˜λ (t1 (vn , λ), vn ) > ( − )λk q cq . p q

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

224

˜λ (t2 (v, λ), v) is We show by the same method that the functional v 7→ E bonded bellow. Put α1 (λ) =

α2 (λ) =

inf

˜λ (t1 (v, λ), v). E

(8)

inf

˜λ (t2 (v, λ), v). E

(9)

v∈Lp (Ω)\{0}

v∈Lp (Ω)\{0}

We have the following lemma: ˆ then Lemma 2.5. If λ ∈ ]0, λ[, α1 (λ) =

inf

v∈S,v≥0

˜λ (t1 (v, λ), v) E

and

α2 (λ) =

inf

v∈S,v≥0

˜λ (t2 (v, λ), v), E

where S is the unit sphere of Lp (Ω). ˜λ (t, v) > 0, then t ∈ ]t1 (v, λ), t2 (v, λ)[. Proof : Let t > 0. If ∂t E Since |Λv| ≤ Λ|v|, we deduce that ˜λ (ti (|v|, λ), |v|) = 0, ∂t E˜λ (ti (|v|, λ), v) ≥ ∂t E

i = 1, 2.

It follows that ]t1 (|v|, λ), t2 (|v|, λ)[ ⊆ ]t1 (v, λ), t2 (v, λ)[. Hence, t1 (|v|, λ) ≥ t1 (v, λ). ˜λ (t, |v|) is decreasing on ]0, t1 (|v|, λ)], we get Using the fact that t 7→ E ˜λ (t1 ((v, λ), |v|) ≥ E ˜λ (t1 (|v|, λ), |v|) E and since |Λv| ≤ Λ|v|, we get ˜λ (t1 (v, λ), |v|). E˜λ (t1 (v, λ), v) ≥ E Hence we conclude that ˜λ (t1 (v, λ), v). E˜λ (t1 (|v|, λ), |v|) ≤ E ˜λ (t, v) is creasing on Since |Λv| ≤ Λ|v| and the function t 7→ E [t1 (v, λ), t2 (v, λ)], we obtain ˜λ (t2 (|v|, λ), |v|) ≤ E˜λ (t2 (|v|, λ), v) E ≤ E˜λ (t2 (v, λ), v). Finally, we have showed that for every v ∈ Lp (Ω) \ {0} ˜λ (ti (|v|, λ), |v|) ≤ E ˜λ (ti (v, λ), v), E

where

i = 1, 2.

(10)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

225

Moreover, for every γ > 0 , we get ˜λ (γt, v ) = E ˜λ (t, v), E γ ˜λ (γt, v ) = 1 ∂t E ˜λ (t, v), ∂t E γ γ ˜λ (γt, v ) = 1 ∂tt E˜λ (t, v). ∂tt E γ γ2 It follows that t1 (v, λ) =

1 v t1 ( , λ), γ γ

(11)

t2 (v, λ) =

1 v t2 ( , λ). γ γ

(12)

By the virtue of (10), (11) and (12), we conclude that α1 (λ) = α2 (λ) =

inf

˜λ (t1 (v, λ), v), E

(13)

inf

˜λ (t2 (v, λ), v), E

(14)

v∈S,v≥0

v∈S,v≥0

where S is the unit sphere of Lp (Ω).

Lemma 2.6. Let (vn ) ⊂ S be a minimizing sequence of (13) (resp. of (14)). Then, (Vn ) := (t1 (vn , λ)vn ) (resp. (Wn ) := (t2 (vn , λ)vn )) are Palais-Smale sequences for the functional Eλ . Proof : We will show this lemma only for the sequence (Vn ), the proof for (Wn ) can be done in the same way. ˆ Then lim Eλ (Vn ) = α1 (λ). Let λ ∈]0, λ[. n→+∞

Now we show that lim Eλ0 (Vn ) = 0. n→+∞

Notice that for every v ∈ Lp (Ω) \ {0}, we have ∂t E˜λ (t1 (v, λ), v) = 0 and ˜λ (t1 (v, λ), v) 6= 0. The implicit function theorem implies that the func∂tt E ˜λ is. Let us introduce the C 1 functional tional v 7→ t1 (v, λ) is C 1 since E f1,λ defined on S by ˜λ (t1 (v, λ), v) = Eλ (t1 (v, λ)v). f1,λ (v) = E Hence α1 (λ) = inf f1,λ (v) = v∈S

inf

v∈S,v≥0

f1,λ (v)

and

lim f1,λ (vn ) = α1 (λ).

n→+∞

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

226

Using the Ekeland variational principle on the complete manifold (S, || ||p ) to the functional f1,λ , we conclude that 1 ||ϕ||p , for every ϕ ∈ Tvn S, n where Tvn S is the tangent space to S at the point vn . ˜λ (t1 (vn , λ), vn ) ≡ 0, then for every ϕ ∈ Tvn S, one has Moreoever, since ∂t E 0 f1,λ (vn )(ϕ) = ∂t E˜λ (t1 (vn , λ), vn )∂v t1 (vn , λ)(ϕ) ˜λ (t1 (vn , λ), vn )(ϕ) +∂v E ˜ = ∂v Eλ (t1 (vn , λ), vn )(ϕ), 0 |f1,λ (vn )(ϕ)| ≤

where ∂v t1 (vn , λ) denotes the derivative of t1 (., λ) with respect to its first variable at the point (vn , λ). Furthermore, let P : Lp (Ω)\{0} → R × S v 7→ (P1 (v), P2 (v)) = (k v kp ,

v ). k v kp

Applying H¨ older’s inequality, we get for every (v, ϕ) ∈ Lp (Ω)\{0}×Lp (Ω) : kP20 (v)(ϕ)kp ≤ 2

kϕkp . kvkp

From lemma 2.3 and by the fact that kVn kp = t(vn , λ), there exists positive constant C such that t1 (vn , λ) ≥ C, ∀n ∈ N.

Hence for every ϕ ∈ Lp (Ω), we obtain ˜λ (P1 (Vn ), P2 (Vn ))P 0 (Vn )(ϕ) |E 0 (Vn )(ϕ)| = |∂t E 1 +∂v E˜λ (P1 (Vn ), P2 (Vn ))P20 (Vn )(ϕ)| ˜λ (t(vn ), vn )P20 (Vn )(ϕ)| = |∂v E 0 = |f1,λ (vn )P20 (Vn )(ϕ)| 1 ≤ k P20 (Vn )(ϕ) kp n 2 k ϕ kp ≤ n C

We easily conclude that 0

lim E 0 (Vn ) = 0 in Lp (Ω).

n→+∞

Remark 2.1. Until now, the minimizing sequences we consider are in S and are nonnegative.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

227

3. Existence results ˆ Then the problem Theorem 3.1. Let 1 < q < p < r < p∗2 and λ ∈ ]0, λ[. (Pλ ) has at least two positive solutions. Proof : We will use the notations of the previous lemmas. Since the sequences (Vn ) and (Wn ) are Palais-Smale for the functional Eλ , ˆ satisfy Palais-Smale condition. it suffices to show that Eλ (0 < λ < λ) By lemma 2.3, we deduce that (Vn ) is bonded in Lp (Ω). Passing if necessary to a subsequence, we get  p  Vn * V1 in L (Ω), ΛV * ΛV1 in X,  n ΛVn → ΛV1 in Lr (Ω),

(15) (and in Lq (Ω)).

On the other hand we have,

Z Nq (ΛVn )(ΛVn − ΛV1 )dx hNp (Vn ), Vn − V1 i = h Eλ0 (Vn ), Vn − V1 i + λ Ω Z Nr (ΛVn )(ΛVn − ΛV )dx. + Ω

Eλ0 (Vn )

→ 0, Nq (ΛVn ) → Nq (ΛV1 ) and Nr (ΛVn ) → Nr (ΛV1 ). Moreover, Then hNp (Vn ), Vn − V1 i → 0. The fact that Np is (S+) type implies that Vn → V1 in Lp (Ω). We know that for any minimizing sequence (vn ) of (13), there is a subsequence still denoted by (vn ) such that Vn = t1 (vn , λ)vn and t1 (vn , λ) = ||Vn ||p . Hence t1 (vn , λ) → ||V1 ||p = t1 , which implies that vn → V1 /t1 = v1 ,

and t1 = t1 (v1 , λ),

where v1 ∈ S. In the same way, for any minimizing sequence (vn ) ⊂ S of (14), passing if necessary to a subsequence, there is t2 ∈ ]0, +∞[ such that t2 (vn , λ)vn → t2 in R, vn → v2 = V2 /t2 , where V2 is the limit of the sequence (Wn ) := (t2 (vn , λ)vn ) in Lp (Ω) and t2 = ||V2 ||p = t2 (v2 , λ).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

228

At this stage, it is easy to see that V1 6= V2 . Indeed, since ˜ ˜λ (t2 (v2 , λ), v2 ) < 0, it follows that ∂tt Eλ (t1 (v1 , λ), v1 ) > 0 and ∂tt E ∂tt Eλ (t1 , V1 /t1 ) > 0 and ∂tt Eλ (t2 , V2 /t2 ) < 0. This achieves the proof. ˆ will be deIn the sequel the solutions V1 and V2 of (Pλ0 ), for λ ∈]0, λ[, noted by V1,λ and V2,λ . Also, t1,λ , t2,λ , v1,λ and v2,λ will stand for t1 (v1 , λ), t2 (v2 , λ), v1 and v2 respectively. Theorem 3.2. Let 1 < q 0 for λ ∈ ]0, λ0 [, ˆ Eλ (V2,λ ) < 0 for λ ∈ ]λ0 , λ[,

where λ0 =

r−q q r r−p ˆ λ. ( ) r p

˜λ (t1,λ , v1,λ ) = 0 and ∂tt E˜λ (t1,λ , v1,λ ) > Proof : (i) Let us recall that ∂t E 0. Then P (V1,λ ) − λQ(V1,λ ) − R(V1,λ ) = 0, (p − 1)P (V1,λ ) − λ(q − 1)Q(V1,λ ) − (r −1)R(V1,λ ) > 0. Using the fact that 1 < q 0. Hence r−p q−p Q(V1,λ ) + R(V1,λ ) pq pr −1 ≤ (λ(p − q)Q(V1,λ ) + (p − r)R(v1,λ )) pq < 0.

Eλ (V1,λ ) = λ

(ii) Let v be an arbitrary element of Lp (Ω) \ {0} and let us write ˜λ (t, v) = tq G ˜ λ (t, v), E

where

p−q λ tr−q ˜ λ (t, v) = t P (v) − Q(v) − R(v). G p q r

It follows that ˜λ (t, v) = qtq−1 G ˜ λ (t, v) + tq ∂ G ˜ λ (t, v), ∂t E

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

229

with ˜ λ (t, v) = tp−q−1 ( ∂t G

p−q r − q r−p P (v) − t R(v)). p r

˜ λ (t, v) is increasing on It is clear that the real valued function t → G ]0, t0 (v)[, decreasing on ]t0 (v), +∞[ and attains its unique maximum for t = t0 (v), where r 1 t0 (v) = ( ) r−p t(v), (16) p and t(v) is defined by the relation (2). On the other hand, a direct computation gives r−q r−q ˜ λ (t0 (v), v) = 1 ( r ) r−p r − p ( p − q P (v) ) r−p R(v) − λ Q(v) . G r p p − q r − q R(v) q

˜ λ (t0 (v), v) > 0 (resp. Similarly, G λ > λ0 (v)) and ˜ λ (v) (t0 (v), v) = 0, where G 0

λ0 (v) =

˜ λ (t0 (v), v) < 0) if λ < λ0 (v) (resp. G

r−q q r r−p ( ) λ(v), r p

with λ(v) given by (3). Thus, we get ˜  Eλ (t0 (v), v) > 0 if λ < λ0 (v), E˜ (t (v), v) = 0 if λ = λ0 (v),  ˜λ 0 Eλ (t0 (v), v) < 0 if λ > λ0 (v).

(17)

(18)

First, since the function

]0, 1[ → R ln t t→ 1−t

is increasing, then for every real numbers x and y such that 0 < x < y, one has 1−x 1 1 1−x 1 ln( ) = ln( ) 1−y , ln( ) > x 1−y y y and consequently

1−x

0 < x(1/y) 1−y < 1. q p In the particular case x = and y = , we get r r r−q q r r−p 0< ( ) < 1, r p

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

230

and therfore 0 < λ0 (v) < λ(v). ˜ λ (v) (t, v) < 0 for t ∈ Moreover, for every v ∈ Lp (Ω) \ {0}, one has G 0 ˜ ]0, +∞[\{t0 (v)} and Gλ0 (v) (t0 (v), v) = 0. Hence, the real valued function ˜λ (v) (t, v), (t > 0), attains its unique maximum at t = t0 (v) and we t→E 0 obtain the following interesting identity t2 (v, λ0 (v)) = t0 (v).

(19)

On the other hand, let λ0 =

inf

v∈Lp (Ω)\{0}

λ0 (v).

(20)

(3) and (16) implies that r−q

P r−p (v) p r r−q . λ0 (v) = ( ) r−p cˆ p−q q p Q(v)R r−p (v) Let us put p−q

M = {v ∈ Lp (Ω), Q(v)R r−p (v) = 1}. It is clair that M is weakly closed. r−q Moreover, the functional v 7→ P r−p (v) is weakly lower semi-continuous and coercive on M . Thus this functional attaints its minimum on M . The p−q r−q homogeneities of v 7→ P r−p (v) and v 7→ Q(v)R r−p (v) enables us to conclude that there is v ∗ ∈ S such that inf λ0 (v) =

v∈M

inf

v∈Lp (Ω)\{0}

λ0 (v) = inf λ0 (v) = λ0 (v ∗ ) = λ0 . v∈S

Now, let λ ∈ ]0, λ0 [, Then, for every v ∈ Lp (Ω) \ {0} one has λ < λ0 (v) ˜λ (t0 (v), v) > 0 holds from (18). Then the function t 7→ and consequently, E ˜ ˜λ (t2 (v, λ), v) > Eλ (t, v), (t > 0) attains its maximum at t2 (v, λ) such that E p ˜ 0 for every v ∈ L (Ω) \ {0}. In particular, we have Eλ (t2 (v2,λ , λ), v2,λ ) > 0, i.e. Eλ (V2,λ ) > 0. If λ = λ0 , then ˜λ0 (t2 (v2,λ0 ), v2,λ0 ) Eλ0 (V2,λ0 ) = E ˜λ0 (t2 (v, λ0 ), v) = inf E v∈S ˜λ0 (t2 (v ∗ , λ0 (v ∗ )), v ∗ ) ≤E ˜λ (v∗ ) (t0 (v ∗ ), v ∗ ) =E 0

= 0, which implies that Eλ0 (V2,λ0 ) ≤ 0. In addition, it is known from ˜λ0 (t0 (v), v) ≥ 0, for every v ∈ Lp (Ω) \ {0}. Then, (18) that E

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

231

˜λ0 (t2 (v2,λ0 , λ0 ), v2,λ0 ) is a global maximum of the function t 7→ since E ˜λ0 (t, v2,λ0 ), (t > 0), we have E ˜λ0 (t2 (v2,λ0 , λ0 ), v2,λ0 ) ≥ E ˜λ0 (t0 (v2,λ0 ), v2,λ0 ) ≥ 0. E We conclude that ˜λ0 (t2 (v2,λ0 , λ0 ), v2,λ0 ) = 0. Eλ0 (V2,λ0 ) = E ˆ Finally, suppose that λ0 < λ < λ. We know that for every (t, v) ∈ ]0, +∞[×Lp(Ω) \ {0}, the real valued ˆ hence we deduce ˜λ (t, v) is decreasing on [λ0 , λ], function λ 7→ E ˜λ (t2 (v2,λ , λ), v2,λ ) = inf E ˜λ (t2 (v, λ), v) E v∈S ≤ E˜λ (t2 (v ∗ , λ), v ∗ ) < E˜λ0 (t2 (v, λ), v).

Moreover, the real valued function t 7→ E˜λ0 (t, v ∗ ), (t > 0), attains its unique maximum for t = t0 (v ∗ ).Then ˜λ0 (t2 (v ∗ , λ), v ∗ ) ≤ E = =

˜λ0 (t0 (v ∗ ), v ∗ ) E ˜λ (v∗ ) (t0 (v ∗ ), v ∗ ) E 0

0.

˜λ (t2 (v2,λ , λ), v2,λ ) < 0, which achieves this proof. Hence E

Theorem 3.3. if v ∗ is a solution of (20), then t0 (v ∗ )v ∗ is a solution of (Pλ0 0 ). Proof : Let v ∗ be a solution of (20), then λ0 = λ0 (v ∗ ) and for every h ∈ Lp (Ω), we have Eλ0 0 (t0 (v ∗ )v ∗ )(h) =

=

λ0 1 p−1 t0 (v)hP 0 (v), hi − tq−1 (v)hQ0 (v), hi p q 0 1 − t0r−1 (v)hR0 (v), hi r P (v)(t0 (v))p−1 hP 0 (v), hi ( p P (v) 0 pλ0 q−p hQ (v), hi p r−p hR0 (v), hi − t − t0 ). q 0 P (v) r P (v)

By the virtue of relations (2), (3), (16) and (17), a direct computation gives for every h ∈ Lp (Ω) pλ0 q−p hQ0 (v ∗ ), hi r − p hQ0 (v), hi t0 = , q P (v ∗ ) r − q Q(v ∗ )

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

232

and p r−p hR0 (v ∗ ), hi p − q hR0 (v), hi t0 ) = . r P (v ∗ ) r − q R(v ∗ ) Then Eλ0 0 (t0 (v ∗ )v ∗ )(h) = K(

r − q hP 0 (v ∗ ), hi hQ0 (v ∗ ), hi p − q hR0 (v ∗ ), hi − − ), r − p P (v ∗ ) Q(v ∗ ) r − p R(v ∗ )

where K=

r − p P (v ∗ ) [t0 (v ∗ )]p−1 . r−q p

On the other hand, the relations (3) and (17) implies that for every h ∈ Lp (Ω) hλ00 (v ∗ ), hi = λ0 (v ∗ )(

r − q hP 0 (v ∗ ), hi hQ0 (v ∗ ), hi p − q hR0 (v ∗ ), hi − − ). r − p P (v ∗ ) Q(v ∗ ) r − p R(v ∗ )

Since hλ00 (v ∗ ), hi = 0 for every h ∈ Lp (Ω), we deduce that hEλ0 0 (t0 (v ∗ )v ∗ ), hi =

K 0 ∗ hλ (v ), hi = 0, λ0 0

for every h ∈ Lp (Ω). Which implies that t0 (v ∗ )v ∗ is a solution of (Pλ0 0 ). Remark 3.1. It is very interesting to notice that in the case of homogeneous Dirichlet boundary condition, we have R p Ω |v(x)| dx ˆ R , inf lim λ = q→p |Λv(x)|p dx v∈Lp (Ω)\{0} Ω

ˆ is the first eigenvalue of the problem Hence, in the case where p = q, λ 0 0 ˆ and has no (Pλ ), i.e. the problem (Pλ ) has positive solutions for λ ∈ ]0, λ] ˆ positive solution for λ > λ. References 1. R. A. Adams, Sobolev Spaces, Academic Press, New York, (1975). 2. C. O. Alves, Existence of Positive Solutions for a Problem with Lack of Compactness Involving the p-Laplacian, Nonlinear Analysis- TMA, 51 (2002), 1187-1206. ˜ o and O. H. Miyagaki, Signed solution for 3. C. O. Alves, P. C. Carria a class of quasilinear elliptic problem with critical growth, Commun. Pure Appl. Anal. 1 (2002), no. 4, 531-545.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

233

4. A. Ambrosetti, H. Brezis and G. Cerami, Combined effects of concave and convex nonlinearities in some elliptic problems, J. Funct. Anal. 122 (1994), no. 2, 519-543. 5. A. Ambrosetti, J. Garcia Azorero and I.Peral, Multiplicity results for some nonlinear elliptic equations, J. Funct. Anal. 137 (1996) no. 1, 219-242. 6. A. Ambrosetti, J. Garcia Azorero and I.Peral, Existence and multiplicity results for some nonlinear elliptic equations: a survey, Rend. Mat. Appl. (7) 20 (2000), 167-198. 7. T. Bartsch and M. Willem, On an elliptic equation with concave and convex nonlinearities, Proc. Amer. Math. Soc. 123 (1995), no. 11, 3555-3561. 8. J.Berkovits and V.Mustonen, Nonlinear mappings of monotone type (classification and degree theory), 1988, Math. Univer. Oulu, Linnanmaa, Oulu, Finland. 9. L. Boccardo, M. Escobedo and I. Peral, A Dirichlet problem involving critical exponents, Nonlinear Anal. 24 (1995), 1639-1648. 10. J. F. Bonder and J. D. Rossi, Existence results for the p-Laplacian with nonlinear boundary conditions, J. Math. Anal. Appl. 263 (2001) 195-223. 11. J. F. Bonder and J. D. Rossi, Asymptotic behavior of the best Sobolev trace constant in expanding and contracting domains, Commun. Pure Appl. Anal. 1 (2002), no. 3, 359-378. ˆ ` bek and M. Otani, 12. P. Dra Global Bifurcation Result for the p-Biharmonic Operator, Electronic Journal Differential Equations 2001(2001), No 48, pp. 1-19. ` bek and S. Pohozaev, Positive Solutions for the p-Laplacian: Appli13. P. Dra cation of the fibering method, Proc. Roy. Soc. Edinburgh Sect. A 127 (1997) 703-726. 14. I. Ekeland, On the Variational Principle, J. Math. Anal. Appl. 47 (1974) 324-353. 15. A. El Hamidi, Multiple Solutions with Changing Sign Energy to a Nonlinear Elliptic Equations, Comm. Pure Appl. Anal. 3 (2004) 253-265. 16. D. Gilbar and N. S. Trudinger, Elliptic Partial Differential Equations of Second Order, Second ed., Springer New York Tokyo (1983). 17. P. Korman, On uniqueness of positive solutions for a class of semilinear equations, Discrete Contin. Dyn. Syst. 8 (2002), no. 4, 865-871. 18. M. Talbi and N. Tsouli, Existence of Solutions for a Nonlinear Elliptic Problem of Fourth Order with Weight, Mediterr. j. math. 3 (2006), 87-96. 19. M. Willem, Minimax Theorems, Progress in Nonlinear Differential Equations and their Applications, 24. Birkhuser Boston, Inc., Boston, MA, (1996).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

234

A survey on potential theory on Orlicz spaces N. A¨ıssaoui D´ epartement de Math´ ematiques Ecole Normale Sup´ erieure B. P. 5206, Ben Souda, F` es Morocco In this survey we expose the fundamentals of the strongly nonlinear potential theory and relate some of its components. As applications, we establish a relation between this theory and Partial Differential Equations, and show whether the equation ∆A u + h = 0 possesses a solution or not, for a fixed function h. Here ∆A is the A-Laplacian which is the p-Laplacian ∆p , when the Orlicz space LA reduces to the Lebesgue space Lp . Keywords: Orlicz spaces; Capacity; Potentials.

1. Introduction The nonlinear potential theory on Lebesgue spaces studied by different schools has introduced the notion of capacity in these spaces and has permitted very rich applications in functional analysis, in harmonic analysis, in the theory of probabilities, in the partial differential equations theory, and so forth. This theory has taken a considerable time for a development in different directions and in different schools. It is one of the most complete theories. Its results permitted, among others, to deepen our knowledge of Sobolev spaces and to solve a great number of problems in higher mathematics. Currently, it makes intervene more general spaces that Sobolev spaces, like Besov spaces and Lizorkin-Triebel spaces. The birth of this theory goes up again to years seventy, to the works of V. G. Maz’ya [75], [76], J. Serrin [89], [90], Yu. G. Reshetnyak [86], and B. Fugled [49], [50], [51], [52], [53]. It took a new thrust during 1970’s with notably the works of B. Fugled [54], [55], [56], [57], N. G. Meyers [80], [81], [82], [83], [84], L. I. Hedberg [63], [64], [65], [66], [67], [68], V. G. Maz’ya [77], [78], Yu. G. Reshetnyak [87], T. Sj¨ odin [91], D. R. Adams [1], [2], [3], [4], [5], [6], [7], D. R. Adams and N. G. Meyers [18], [19], D. R. Adams and J. C. Polking [21], V. P. Havin [60],

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

235

[61], V. P. Havin and V. G. Maz’ya [62], C. Fernstr¨ om [48], K. Hansson [59] and others. Other components and applications of this theory are treated in different papers. See for example D. R. Adams [8], [9], [10], [11], [12], [13], D. R. Adams and A. Heard [14], D. R. Adams and L. I. Hedberg [15], D. R. Adams and J. L. Lewis [17], D. R. Adams and M. Pierre [20], H. Aikawa [24], [25], [26], [27], B. Fugled [54], L. I. Hedberg [69], L. I. Hedberg and Th. H. Wolff [70], B. Jawerth, C. P´erez, G. Welland [71], B. O. Turesson [95] and others. The interested reader can consult the indispensable books by D. R. Adams and L. I. Hedberg [16], by W. P. Ziemer [96], by E. M. Stein [92] and by V. G. Maz’ya [79]. We are sure that many important references are missing, because the number of publications is very important, so it is impossible to mention them. The increasing necessity to work on other spaces to solve other types of equations, says strongly non linears, motivated the creation of a strongly non linear potential theory, that we have introduced in different works. An embryo of this theory is in the thesis of A. Benkirane [46] and in the paper [47] by A. Benkirane and J. P. Gossez. This new theory uses Orlicz spaces and Sobolev-Orlicz spaces that are natural generalizations of Lebesgue and Sobolev spaces. It is important to notice that the main results of the non linear theory spread to the case of strongly non linear one. This new theory doesn’t stop marking progress. In this survey we mention some components of this theory, namely 1) the introduction and the study of a capacity in Orlicz spaces, the capacitability of analytic sets and a potential in Orlicz spaces, 2) the study of the continuity of the potential and in particular, the Bessel potential in Orlicz spaces, 3) the study of the instability phenomena of the capacity in Orlicz spaces, 4) the study of some relations between the capacity and maximal operators in Orlicz spaces, 5) the introduction and the study of Wolff inequality in strongly nonlinear potential theory and applications, 6) the introduction and the study of the notion of quasicontinuity in Orlicz spaces. The first relation between the Strongly Nonlinear Potential Theory and Partial Differential Equations is the fact that a compact set K is removable for an elliptic linear operator of order m, with constant coefficients, if and only if its Bessel capacity is null (i.e. Bm,A (K) = 0). On the other hand we know that an important application of the non

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

236

linear potential theory is the resolution of some equations involving the pLaplacian operator. Hence the p-Laplace equation ∆p u+h = 0 on IRN , N ≤ p, has no solution if h has a non zero average and the equation ∆p u+g = 0, on a p-hyperbolic manifold M , has a solution with p-integrable gradient for any bounded measurable function g : M → IR with compact support. In [37] we establish that if the N -function A satisfies the ∆2 condition and IRN is A-parabolic, then the equation ∆A u + h = 0 has no weak solution for any function h having a non zero average. Here ∆A is the A-Laplacian which is the p-Laplacian ∆p , when the Orlicz space LA is the Lebesgue space Lp . We establish also, for a large class of Orlicz spaces LA including Lebesgue spaces Lp (p > 1), that if the function h is in L∞ and has a compact support, then the equation ∆A u + h = 0 has a weak solution when IRN is A-hyperbolic. These results generalize those of V. Gol’dshtein, M. Troyanov in [55] and of M. Troyanov in [91]. We can’t develop here other components like the boundedness principle, the weighted strongly nonlinear potential theory or the strongly nonlinear potential theory on metric spaces, because they are out of the scope of this survey. The interested reader can consult [33], [36], ]38], [40], [41]. 2. Preliminaries 2.1. Orlicz spaces An N-function is a continuous convex and even function A defined on IR, verifying A(t) > 0 for t > 0, A(t) A(t) = 0and lim = +∞. t→+∞ t→0 t t lim

We have the representation A(t) =

R|t| 0

a(x)dx, where a : IR+ → IR+ is

non-decreasing, right continuous, with a(0) = 0, a(t) > 0 for t > 0 and limt→+∞ a(t) = +∞. R|t| ∗ The N -function A∗ conjugate to A is defined by A∗ (t) = a (x)dx, 0

where a∗ is given by a∗ (s) = sup{t : a(t) ≤ s}. Let A be an N -function and Ω an open set in IRN . We note LA (Ω) the set, called an Orlicz class, of measurable functions f , on Ω, such that R ρ(f, A, Ω) = Ω A(f (x))dx < ∞.

Let A and A∗ be two conjugate N -functions and let f be a measurable

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

237

function defined almost everywhere in Ω. The Orlicz norm of f , ||f ||A,Ω or ||f ||A if there is no confusion, is defined by R ||f ||A = sup{ Ω |f (x)g(x)|dx : g ∈ LA∗ (Ω), and ρ(g, A∗ , Ω) ≤ 1}. The set LA (Ω) of measurable functions f such that ||f ||A < ∞, is called an Orlicz space. When Ω = IRN , we set LA in place of LA (IRN ). The Luxemburg norm |||f |||A,Ω or |||f |||A if there is no confusion, is defined in LA (Ω) by o n R dx ≤ 1 . |||f |||A = inf r > 0 : Ω A f (x) r Orlicz and Luxemburg norms are equivalent. More precisely, if f ∈ LA (Ω), then |||f |||A ≤ ||f ||A ≤ 2|||f |||A . Let A be an N -function. We say that A verifies the ∆2 condition if there is a constant C > 0 such that A(2t) ≤ CA(t) for all t ≥ 0. Recall that A verifies the ∆2 condition if and only if LA = LA . Moreover LA is reflexive if and only if A and A∗ verify the ∆2 condition. We recall the following results. Let A be an N -function and a its derivative. Then 1) A verifies the ∆2 condition if and only if one of the following holds i) ∀r > 1, ∃k = k(r) : (∀t ≥ 0, A(rt) ≤ kA(t)) ; ii) ∃α > 1 : (∀t ≥ 0, ta(t) ≤ αA(t) ) ; iii)∃β > 1 : (∀t ≥ 0, ta∗ (t) ≥ βA∗ (t) ) ; ∗ 0 ∗ iv) ∃d > 0 : ∀t ≥ 0, A t(t) ≥ d a t(t) .

Moreover α in ii) and β in iii) can be chosen such that α−1 + β −1 = 1. We note α(A) the smallest α such that ii) holds. 2) If A verifies the ∆2 condition, then i) ∀t ≥ 1, A(t) ≤ A(1)tα and ∀t ≤ 1, A(t) ≥ A(1)tα ii) ∀t ≥ 1, A∗ (t) ≥ A∗ (1)tβ and ∀t ≤ 1, A∗ (t) ≤ A∗ (1)tβ . See for instance [47], [73] and [88]. Let A be an N -function such that A and A∗ satisfy the ∆2 condition. We note α(A) = α and α(A∗ ) = α∗ . Then we have from 2) above ∀t ≥ 0, α∗ A∗ (t) ≥ ta∗ (t). Hence β ≤ α∗ . If β = α∗ , then ∀t ≥ 0, α∗ A∗ (t) = ta∗ (t).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

238

This implies that there exists a constant C, such that: ∀t ≥ 0, A∗ (t) = Ct . This means that we are in the case of Lebesgue classes Lp , which is treated in the literature. Hence we suppose in the sequel that β < α∗ . b be an N -function equal to A in a Let A be an N -function and let A R 1 b−1 neighborhood of infinity and such that (see [23, lemma 4.4]): 0 A1+ (t) 1 dt < t N ∞ R b−1 b ∞. If A1+ (t) 1 dt = ∞, we define a new N -function A1 by the formula α∗

1

t

N

b−1 (x) = A 1

Rx Ab−1 (t) 1

0

t1+ N

dt

and we let A1 to be an N -function equal to A in a neighborhood of 0 and to b1 in a neighborhood of infinity (see [23, lemma 4.5] for the construction of A R∞ A−1 1 (t) such N -function). If dt = ∞, we start again the same construction 1+ 1 1

t

N

and we put A2 = (A1 )1 , ... .

Let j = J(A, N ) be the smallest integer such that If

R∞ A−1 (t) 1

0

t1+ N

R∞ A−1 j (t) 1

1

dt < ∞, we put J(A, N ) = 0.

t1+ N

dt < ∞ .

Observe that J(A, N ) ≤ N because there exists a constant C, such that A−1 (t) ≤ Ct, ∀t ≥ 1. Let m be a positive integer. The Orlicz-Sobolev space W m LA (Ω) is the space of real functions f , such that f and its distributional derivatives up to order m, are in LA (Ω). The space W m LA (Ω) is a Banach space equipped with the norm: |||f |||m,A =

P

|i|≤m

|||Di f |||A , f ∈ W m LA (Ω) .

Let W −m LA∗ (Ω) denote the space of distributions on Ω, which can be written as sums of derivatives up to order m of functions in LA∗ (Ω). It is a Banach space under the usual quotient norm. ∗ Recall that satisfy the ∆2 condition, the dual of if A and A −m m N W LA IR coincides with W LA∗ IRN . For more details on the theory of Orlicz spaces, see [22[, [72], [73], [74] and [88].

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

239

2.2. Capacity and Bessel potentials We define a capacity as an increasing positive set function C given on a σadditive class of sets Γ, which contains compact sets and such that C(∅) = 0 S P and C( Xi ) ≤ C(Xi ) for Xi ∈ Γ, i = 1, 2, ... . i≥1

i≥1

C is called outer capacity if for every X ∈ Γ,

C(X) = inf {C(O) : O open, X ⊂ O} . Let k be a positive and measurable function on IRN and let A be an N -function. For X ⊂ RN , we define 0 Ck,A (X) = inf{|||f |||A : f ∈ L+ A and k ∗ f ≥ 1 on X},

and 0 Ck,A (X) = A Ck,A (X) ,

where k∗ f is the usual convolution. The sign + deals with positive elements in the considered space. If a statement holds except on a set X where Ck,A (X) = 0, then we say that the statement holds Ck,A -quasieverywhere (abbreviated Ck,A − q.e or (k, A) − q.e if there is no confusion). We call a function f in L+ A such that k ∗ f ≥ 1 on X, a test function for 0 0 0 Ck,A (X). Moreover, a test function, say f , for Ck,A (X) such that Ck,A (X) = 0 |||f |||A is called a Ck,A −capacitary distribution for X and k ∗ f is called a 0 Ck,A −capacitary potential for X. M denotes the vector space of Radon measures. M1 is the Banach space of measures equipped with the norm ||µ|| = total variation of µ < ∞. The space of measures supported by a compact K is denoted by M(K), and the cone of positive elements is M+ (K). F will stand for the σ−field of sets which are µ−measurable for all µ ∈ M+ 1. If µ ∈ M+ 1 , we say that µ is concentrated on X if µ(y) = 0 for all sets Y which are µ−measurable and such that Y ⊂ c X. Let A and A∗ be two conjugate N -functions. For X ∈ F, we define Dk,A (X) = , µ concentrated on X and ||k ∗ µ||A∗ ≤ 1 sup ||µ|| : µ ∈ M+ 1 R where k ∗ µ is the convolution of k and µ defined by (k ∗ µ)(x) = k(x − y)dµ(y).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

240

A measure µ ∈ M+ 1 such that µ is concentrated on X and ||k ∗ µ||A∗ ≤ 1 is called a test measure for Dk,A (X). If in addition Dk,A (X) = ||µ||, we say that µ is a Dk,A −capacitary distribution for X and k ∗ µ is called a Dk,A −capacitary potential for X. For m > 0, the Bessel kernel Gm is defined through its Fourier trans − m2 −N 2 form IF (Gm ) as [IF (Gm )] (x) = (2π) 2 1 + |x| where [IF (f )] (x) = R −N (2π) 2 f (y)e−ixy dy for f ∈ L1 . Gm is positive, in L1 and verifies the equality: Gr+s = Gr ∗ Gs . For more details on Bessel kernels, see [44], [45] and [92]. m−N We note Im (x) = |x| the Riesz kernel. 0 0 0 We put Bm,A = CGm ,A , Bm,A = CG , Rm,A = CIm ,A and Rm,A = m ,A 0 CIm ,A . 3. Capacity and non linear potential in Orlicz spaces The essential results concerning the notion of capacity in Orlicz spaces are resumed in the following theorem; see [28] and [42]. Theorem 3.1. Let A be an N -function. Then 0 a) Ck,A is an outer capacity defined on all subsets of IR N . 0 b) Ck,A (X) = 0 if and only if there exists f ∈ L+ A such that k ∗ f = +∞. c) If fn → f strongly in LA , then

0 (1) k ∗ fn → k ∗ f in Ck,A −capacity, (2) there exists a subsequence (f ”n )n of the sequence (fn )n such that 0 k ∗ f ”n → k ∗ f Ck,A -quasi uniformly, 0 (3) k ∗ f ”n → k ∗ f Ck,A -quasi everywhere.

d) If (Kn )n is a decreasing sequence of compact set and K = ∩n Kn , then 0 0 lim Ck,A (Kn ) = Ck,A (Kn ).

n→∞

e) If (Xn )n is a increasing sequence of sets in IRN and if LA is a reflexive 0 0 Orlicz space, then limn→∞ Ck,A (Xn ) = Ck,A (∪n Xn ). 0 -capacitable. f ) If LA is a reflexive Orlicz space, then analytic sets are Ck,A ∗ g) If A is an N -function and Dk,A is the outer capacity associated to Dk,A , ∗ 0 (X). (X) = Ck,A then for any set X, Dk,A h) If LA is a reflexive Orlicz space, then for all analytic sets X 0 Dk,A (X) = Ck,A (X).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

241

Let A be an N -function and k a positive and measurable function defined 0 on IRN . Let X be any set such that Ck,A (X) < ∞. We consider the following + 0 variational problem: Find f0 ∈ LA such that k ∗ f0 ≥ 1 Ck,A -q.e. on X and + 0 |||f0 |||A = inf{|||f |||A : f ∈ LA , k ∗ f ≥ 1 Ck,A -q.e. on X}. An answer is given in the following theorem; see [43]. Theorem 3.2. 1) Let LA be a reflexive Orlicz space and X be any set such 0 0 that Ck,A (X) < ∞. Then X has a unique Ck,A -capacitary distribution f ; + 0 f ∈ LA , k ∗ f ≥ 1 on X and Ck,A (X) = |||f |||A . 2) Let LA be a reflexive Orlicz space and X be an analytic set. If γ 0 is the Dk,A −capacitary distribution for X and f is the Ck,A -capacitary distribution for X, then

−1

a ◦ (f .|||f |||−1 k ∗ γ = a ◦ (f .|||f |||−1 A ) a.e. A ) A∗ Moreover k ∗ f ≤ 1 on suppγ, where a is the derivative of A.

4. On the Bessel potentials in Orlicz spaces 4.1. On the continuity of Bessel potentials in Orlicz spaces The results in this subsection show that Bessel capacities in reflexive Orlicz spaces are non increasing under orthogonal projection of sets; see [29]. This is used to get a continuity of potentials on some subspaces. The obtained results generalize those of Meyers and Reshetnyak in the case of Lebesgue classes. Theorem 4.1. 1) Let A be an N -function such that A and A∗ satisfy the ∆2 condition. Let k be a kernel on IRN which is spherically symmetric and non-increasing as |x| increases. If S is an affine subspace of IR N and X a subspace of IRN , then Ck,A (PS X) ≤ Ck,A (X) .

2) Let A be an N -function such that A and A∗ satisfy the ∆2 condition. Let k be a kernel on IRN which is spherically symmetric and nonincreasing as |x| increases. Further, suppose that k is locally integrable with lim|x|→∞ k(x) = 0. Let S be an affine subspace of IRN . Then a) For f ∈ LA and ε > 0, there exists a closed set F ⊂ S such that Ck,A (S − F ) < ε and k ∗ f ∈ C0 (F + S ⊥ ).

Hence k ∗ f ∈ C0 (x + S ⊥ ) Ck,A − q.e. in S.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

242

b) Let (fi )i be a sequence convergent to f in LA . Then there is a subsequence (fi0 )i0 , such that given ε > 0, there exists a closed set F ⊂ S with the property Ck,A (S − F ) < ε and k ∗ fi0 → k ∗ f in C0 (F + S ⊥ ). Hence k ∗ fi0 → k ∗ f in C0 (x + S ⊥ ) Ck,A − q.e. in S. Theorem 4.2. Let A be an N -function such that A and A∗ satisfy the ∆2 condition. Let S be an affine subspace of IRN and 0 ≤ s < m. Then (1) For f ∈ LA and ε > 0, there exists a closed set F ⊂ S such that Bm−s,A (S − F ) < ε and Dj (Gm ∗ f ) ∈ C0 (F + S ⊥ ) for all j, |j| ≤ s. Hence for such j, Dj (Gm ∗ f ) ∈ C0 (x + S ⊥ ) Bm−s,A − q.e. in S. (2) Let (fi )i be a sequence convergent to f in LA . Then there is a subsequence (fi0 )i0 , such that given ε > 0, there exists a closed set F ⊂ S with the property Bm−s,A (S − F ) < ε and Dj (Gm ∗ fi0 ) → Dj (Gm ∗ f ) in C0 (F + S ⊥ ) for all j, |j| ≤ s. Hence for such j, Dj (Gm ∗ fi0 ) → Dj (Gm ∗ f ) in C0 (x + S ⊥ ) Bm−s,A − q.e. in S. Theorem 4.3. Let A be an N -function and T be a one to one map of IR N onto itself. Suppose that T and its inverse T −1 satisfy a Lipschitz condition. Let ρ, 0 < ρ < ∞, and X ⊂ IRN be such that diam X ≤ ρ. Then there exists a constant C, independent of X such that 0 0 Bm,A [T (X)] ≤ CBm,A (X).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

243

If IRN is the affine direct sum of G and H, we define PGH as the projection of IRN onto G, parallel to H. Theorem 4.4. Let A be an N -function such that A and A∗ satisfy the ∆2 condition. Let ρ, 0 < ρ < ∞ and X ⊂ IRN be such that diam PGH X ≤ ρ. Then there exists a constant C, independent of X such that 0 0 Bm,A (PGH X) ≤ CBm,A (X).

4.2. Instability of capacity in Orlicz spaces In this subsection we describe the instability of certain capacities in Orlicz spaces. Our results generalize some of those obtained by C. Fernstr¨ om in [48] and the Theorem 9 of Hedberg in [7] in the case of Lebesgue spaces. See [30]. m−N For m > 0 we define the Riesz kernel, Im , by Im (x) = |x| for x ∈ IRN . We suppose that m < N. If A is an N -function and X any subset of 0 IRN , we define Rm,A (X) as 0 (X) = inf{|||f |||A : f ∈ L+ Rm,A A and Im ∗ f ≥ 1 on X}. 0 We know that Rm,A is an outer capacity; see [40]. We let Rm,A = 0 . A ◦ Rm,A

Definition 4.1. Let X be a Borel set and m be the Lebesgue measure on IRN . Then x is a density point for X if limr→0 m(B(x, r))

−1

m(X ∩ B(x, r)) = 1.

Theorem 4.5. Let A be an N -function verifying the ∆2 condition. Suppose that m < N α . Let X be a Borel set and x a density point for X. Then 0 −1 0 lim Rm,A (B(x, r)) .Rm,A (X ∩ B(x, r)) = 1. r→0

Corollary 4.1. Let A be an N -function verifying the ∆2 condition. Suppose that m < N α . Let X be a Borel set. Then 0 −1 0 lim Rm,A (B(x, r)) .Rm,A (X ∩ B(x, r)) = 1, a.e. on X. r→0

Theorem 4.6. Let A be an N -function verifying the ∆2 condition and E N be a Borel set. Suppose that m < N one of the following α . Then a.e. on IR relations holds 0 −1 0 lim Rm,A (B(x, r)) .Rm,A (E ∩ B(x, r)) = 1 r→0

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

244

or lim limr−N .Rm,A (E ∩ B(x, r)) = 0.

r→0

Theorem 4.7. Let A be an N -function verifying the ∆2 condition and E be an everywhere dense Borel set. Suppose that m < N α . Then the following are equivalent. (i) Rm,A (E ∩ O) = Rm,A (O) for every open O. (ii) Rm,A (E ∩ B(x, r)) = Rm,A (B(x, r)) for all x and r. (iii) For almost all x (with respect to Lebesgue measure) lim sup r

r→0

−N α

0 Rm,A (E ∩ B(x, r)) > 0.

The next Theorem gives a connection between Riesz and Bessel capacities in the case of Orlicz spaces. Theorem 4.8. (a) Let m be a positive number such that m < N. Then there exists a constant δ, such that for all N -functions A and all sets X ⊂ IRN , 0 0 (X) ≤ δBm,A (X) . Rm,A

(b) Let A be an N -function verifying the ∆2 condition such that m < N α. Let Br be the ball centered at 0 with radius r. Then for all sets X ⊂ Br , there exists a constant δ, independent of X, but dependent on r, such that 0 0 Bm,A (X) ≤ δRm,A (X) .

Remark 4.1. Since A verifies the ∆2 condition, there exists a constant γ 0 , independent of X, but dependent of r, such that Bm,A (X) ≤ γ 0 Rm,A (X) . 4.3. Bessel potentials in Orlicz spaces In this subsection it is shown that Bessel potentials have a representation in term of measure when the underlying space is Orlicz. A comparison between capacities and Lebesgue measure is given and geometric properties of Bessel capacities in this space are studied. Moreover it is shown that if the capacity of a set is null, then the variation of all signed measures of this set is null when these measures are in the dual of an Orlicz-Sobolev space; see [31].

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

245

4.3.1. Representation of potentials and comparison with Lebesgue measure Theorem 4.9. Let A be an N -function such that A and A∗ satisfy the ∆2 condition. Let m be a positive integer and X a set in IR N such that 0 0 < Bm,A (X) < ∞. 0 Let f be the Bm,A −capacitary distribution of X. Then there exists a positive measure µX such that: 0 1) Gm ∗ f = Bm,A (X).Gm ∗ a−1 ◦ (Gm ∗ µX ) , where a is the derivative of A. 2) supp µX ⊂ X. If in addition we suppose that X is compact, then 3) G m ∗ f ≤ 1 on supp µX . Theorem 4.10. Let A be an N -function such that A and A∗ satisfy the ∆2 condition. Let m be a positive integer. Then (1) If m ≤ J(A, N ) there exists a constant C = C(A, N, m) > 0 such that −1 1 0 Bm,A (X) ≥ C A−1 m m∗ (X) for all set X such that m∗ (X) 6= 0. (Here m is the Lebesgue measure on IRN and m∗ is the outer measure associated to m). (2) If m > J(A, N ), there exists a constant C = C(A, N, m) > 0 such that for all set X 6= ∅, 0 (X) ≥ C. Bm,A

Theorem 4.11. (1) Let A be an N -function and m be such that 0 < m < N . Let Sρ = B(x, ρ) be the open ball centered at x and with radius ρ. Then there exists a constant C independent of ρ such that 0 Bm,A (Sρ ) ≤ Cρ−m for 0 < ρ ≤ 1.

(2) Let A be an N -function satisfying the ∆2 condition and let m be such that 0 < m < N . Let C(A) be the smallest constant C 0 such that: A(2t) ≤ C 0 A(t), ∀t. 0 Then there exists a constant C independent of ρ such that: Bm,A (Sρ ) ≤ −q −m C2 ρ for 0 < ρ ≤ 1, where q is the greatest positive integer such Logρ−N that q ≤ LogC(A) .

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

246

4.3.2. Relation between capacity and Hausdorff measure Theorem 4.12. Let A be an N -function such that A and A∗ satisfy the ∆2 condition. Let m be a positive real such that αm < N , where α = α(A). Let u b be a positive, decreasing function defined on IR, continuous from the right and such that Gm (r) ≤ u b(r) and lim u b(r)Gm (r)−1 = +∞. r→0

0

Bub0 ,A ,

0

0 If B = then limρ→0 B (Sρ )Bm,A (Sρ )−1 = 0. 0 In particular, if αm < N , then limρ→0 Bm,A (Sρ ) = 0.

Definition 4.2. Let ϕ(r) be a positive, increasing function in some interval [0, r0 [ and such that limr→0 ϕ(r) = 0. If X is an arbitrary set, the Hausdorff ϕ-measure of X is given by X Hϕ(r) (X) = lim (inf ϕ(ri )), s→0

i≥1

where the above infimum is taken over all countable coverings of X by spheres S(xi , ri ) such that ri ≤ s. Note that Hϕ(r) is a capacity which has the property Hϕ(r) (X) = Hϕ(r) (Y ), where Y is a G-set containing X. Theorem 4.13. Let A be an N -function such that A and A∗ satisfy the ∆2 condition and let X be a subset of IRN . Let m be a positive real such 0 (Sr ). Then that αm < N , where α = α(A) and let ϕ(r) = Bm,A Bm,A (X) = 0 if Hϕ(r) (X) < ∞. 4.3.3. Capacities and measures in Orlicz-Sobolev spaces Theorem 4.14. 1) Let A be an N -function such that A and A∗ satisfy the ∆2 condition and let integer such that m ≤ J(A, N ). m be a positive Let T ∈ W −m LA∗ IRN ∩ M1 IRN and let K be a compact set such that Bm,A (K) = 0 and T − (K) = 0. Then kT k (K) = 0. 2) Let A be an N -function such that A and A∗ satisfy the ∆2 condition and let m be a positive integer such that m ≤ J(A, N ). Let T ∈ W −m LA∗ IRN ∩ M1 IRN and let X be a kT k-measurable set such that Bm,A (X) = 0 and T − (X) = 0.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

247

Then kT k (X) = 0. 3) Let A be an N -function such that A and A∗ satisfy the ∆2 condition and let m be a positive integer such that m ≤ J(A, N ). Let T ∈ W −m LA∗ IRN ∩ M1 IRN and K be a compact set such that Bm,A (K) = 0. Then kT k (K) = 0. 4) Let A be an N -function such that A and A∗ satisfy the ∆2 condition and let m be a positive integer such that m ≤ J(A, N ). Let T ∈ W −m LA∗ IRN ∩ M1 IRN and let X be a set such that Bm,A (X) = 0. Then kT k (X) = 0. 5. Another developments of strongly nonlinear potential theory We establish some relations between potentials and maximal functions in Orlicz spaces. We give a definition of quasicontinuity and obtain a description of quasicontinuous representative in some potential spaces. We also give a result on smooth truncation of potentials in Orlicz-Sobolev spaces and compare some capacities. As a consequence: a compact is removable in Orlicz space for an elliptic linear operator of order m with constant coefficients if and only if its Bessel capacity of order m is null. See [32]. 5.1. Maximal operators and potentials Let f be a locally integrable function. The Hardy-Littlewood maximal function associated to f is defined by M f (x) = M0 f (x) = R −1 supr>0 |B(x, r)| |f (y)| dy. B(x,r)

Here |B(x, r)| is the Lebesgue measure of B(x, r) on IR N . The fractional maximal function associated to f is defined for 0 < α < α−N R N , by Mα f (x) = supr>0 |B(x, r)| N B(x,r) |f (y)|dy. And for 0 ≤ α < N , and δ > 0, the inhomogeneous version of these R α−N |f (y)| dy. functions is defined by Mα,δ f (x) = supδ≥r>0 |B(x, r)| N B(x,r)

For 0 ≤ α < N , and δ > 0, we define the modified Riesz kernel, Iα,δ by Iα,δ (x) = Iα (x), if |x| < δ, Iα,δ (x) = 0, if |x| ≥ δ.

The Riesz potential Iα ∗ µ, 0 < α < N , where µ is a positive measure, can be estimated below by the fractional maximal function associated to µ. In fact, for every r > 0,

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

248

RN

R

α

|x − y| −N dµ(y) ≥

R

|x−y|≤r

|x − y|

α−N

dµ(y) ≥

R

dµ(y).

|x−y|≤r

The reverse inequality is false in general. In the first part of the following theorem, we give a generalization to Orlicz spaces, of the classical theorem of B. Muckenhoupt and R.I. Wheeden [85], which establishes the opposite inequality in term of Lp norms. Theorem 5.1. Let A be an N -function satisfying the ∆2 condition, and let 0 < α < N. Then (1) There is R a constant C R> 0, such that for any R positive measure µ, i) C −1 A(Mα µ)dx ≤ A(Iα ∗ µ)dx ≤ C A(Mα µ)dx ii) C −1 |||Mα µ|||A ≤ |||Iα ∗ µ|||A ≤ C|||Mα µ|||A . (2) If δ is a positive number, there are positive constants C1 , C2 and C3 such that for any positive measure µ, |||Mα,δ µ|||A ≤ C1 |||Iα,δ ∗ µ|||A ≤ C2 |||Gα ∗ µ|||A ≤ C3 |||Mα,δ µ|||A . 5.2. Quasicontinuity We recall the general definition of quasicontinuity. Definition 5.1. Let C be a capacity on IRN and let f be a function defined C−q.e. on IRN or on some open subset of IRN . Then f is said to be C−quasicontinuous if for every > 0, there is an open set O such that C(O) < and f |Oc ∈ C(Oc ). In other words, the restriction of f to the complement of O is continuous in the induced topology. 0 , we write (m, A)-quasicontinuous in place of For Bessel capacity Bm,A

0 Bm,A -quasicontinuous.

Theorem 5.2. 1) Let A be an N -function satisfying the ∆2 condition. If f ∈ LA , then the potential Gm ∗ f , m > 0, is (m, A)−quasicontinuous. Hence every element in Lm,A has an (m, A)−quasicontinuous representative. 2) Let A be any N R -function. Let f = Gm ∗ g ∈ Lm,A , m > 0. Then 1 f (y)dy = Gm ∗ g(x), wherever Gm ∗ |g| (x) < ∞, limr→0 |B(x,r)| B(x,r)

i.e., (m, A) − q.e.

Theorem 5.3. 1) Let A be any N -function. Let f1 and f2 be two (m, A)−quasicontinuous functions, m > 0. Suppose that f1 (x) = f2 (x) almost everywhere. Then

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

249

f1 (x) = f2 (x) (m, A)−quasieverywhere. 2) Let A be an N -function satisfying the ∆2 condition. Let f, g ∈ Lm,A , m > 1. Then D(f g) = f (x)Dg(x) + Df (x)g(x) (m − 1, A) − q.e. We extend the last theorem for any kernel and with a less restrictive property that the quasicontinuity; but for reflexive Orlicz space. This is the tribute to pay! By B(IRN ) we note the family of Borelian sets in IRN . Theorem 5.4. Let k be any kernel and suppose that the N -function A is such that A and A∗ satisfy the ∆2 condition. Let g1 and g2 be two functions verifying the following: ∀ > 0, ∃X ∈ B(IRN ) : Ck,A (X) < and the restrictions of g1 and g2 to c X are continuous. Suppose that {x : g1 (x) 6= g2 (x)} ∈ B(IRN ) and that g1 (x) = g2 (x) a.e. Then g1 (x) = g2 (x)(k, A) − q.e. 5.3. Operations on potentials, Other definition of capacity and removable singularities For reflexive Orlicz spaces, we establish that composition of Bessel potential with a smooth operator, is a potential. This is an extension of a well known Theorem of V.G. Maz’ya which is a substitute of the fact that the Sobolev spaces W m,p (m 6= 1) are not closed under contractions. An immediate consequence is the equivalence of capacities Nm,A and Bm,A . Note that in the case of Lp Lebesgue spaces, this two capacities are equivalent even if m is not integer. See [21]. The correspondent case for Orlicz spaces remains open. On the other hand, we show that in reflexive Orlicz spaces, a compact set K is removable for an elliptic linear operator of order m, with constant coefficients, if and only if its Bessel capacity is null (i.e. Bm,A (K) = 0). This is the first relation between the Strongly Nonlinear Potential Theory, and Partial Differential Equations. Theorem 5.5. Let m be an integer such that 0 < m < N and A be an N -function such that A and A∗ satisfy the ∆2 condition. Let k be an integer k + such that k ≥ m and i−1 T ∈ C (IR ) verifies the following condition (i) sup x T (x) ≤ L < ∞, i = 1, 2, ..., k. Then T ◦ (Gm ∗ f ) ∈ Lm,A , for all f ∈ L+ A , and there is a constant C, which depends only on A, m and N , such that |||T ◦ (Gm ∗ f )|||m,A ≤ CL|||Gm ∗ f |||m,A = CL|||f |||A .

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

250

Definition 5.2. For X ⊂ IRN , we pose Nk,A (X) = inf {A(|||ϕ|||k,A ) : ϕ ∈ S and ϕ = 1 in a neighborhood of X} 0 Nk,A (X) = inf {|||ϕ|||k,A : ϕ ∈ S and ϕ = 1 in a neighborhood of X} . Here S = S(IRN ) is the Schwartz space of rapidly decreasing functions. If k = Gm , we write Nm,A = NGm ,A . Definition 5.3. Let K ⊂ IRN be a compact set, and let P a partial differential operator defined in a neighborhood of K. Then K is said to be removable for P in LA if any solution v of Pv = 0 in O\K for some bounded open neighborhood of K, such that v ∈ LA (O \ K), can be extended to a function ve ∈ LA (O) such that Pe v = 0 in O.

Theorem 5.6. Let m be an integer such that 0 < m < N. Let K ⊂ IR N be a compact set, and let P an elliptic linear partial differential operator of order m with constant coefficients. Let A be an N -function such that A and A∗ satisfy the ∆2 condition. Then K is not removable for P in LA if Bm,A∗ (K) > 0, and it is removable if Nm,A∗ (K) = 0.

We remark the immediate inequality Bm,A (X) ≤ Nm,A (X). In view of the last theorem, it is of considerable interest that these set functions are in fact equivalent. Theorem 5.7. Let m be an integer such that 0 < m < N, and A be an N -function such that A and A∗ satisfy the ∆2 condition. Then there is a constant C such that for all X ⊂ RN , Bm,A (X) ≤ Nm,A (X) ≤ CBm,A (X). This means that a compact K ⊂ IRN , is removable in LA for an elliptic linear operator of order m with constant coefficients if and only if Bm,A (K) = 0. The first Lp version of this theorem was been proved by V.G. Maz’ya [79, Chapter 9.3]. The Lp version for general m is due to D.R. Adams and J.C. Poking [21]. The general case when m ∈ IR is such that 0 < m < N, remains open. 6. Capacitary type estimates in strongly nonlinear potential theory and applications In this section general result on smooth truncation of Riesz and Bessel potentials in Orlicz-Sobolev spaces is given and a capacitary type estimate is presented. We construct also a space of quasicontinuous functions and an alternative characterization of this space and a description of its dual are

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

251

established. For the Riesz kernel Im , we get that operators of strong type (A, A), are also of capacitaries strong and weak types (m, A). See [34]. 6.1. A capacitary type estimate Theorem 6.1. Let A be an N -function such that A and A∗ verify the ∆2 condition, α = α (A) and m is a positive integer. Let Tj∈Z be a doubly infinite sequence of C m (IR) functions identically zero for t < 0 with Tj0 having disjoints supports in (0, ∞) and such that (k) sup tk−1 Tj (t) ≤ L < ∞, k = 0, 1, ..., m. t>0

Then for all f ∈ L+ A , there is a constant C depending only on N, m, L and A such that P |||Dβ Tj (S m ∗ f )|||A ≤ C|||f |||A , j

where β is a multi-index such that |β| = m, and Sm is either Gm if m is a positive integer, or Im if m is a positive integer such that m < N/α. Let S m as above and set S 0m,A (X) = inf{|||f |||A : f ∈ L+ A and S m ∗ f ≥ 1 on X}. Theorem 6.2. Let A be an N -function such that A and A∗ verify the ∆2 condition and m a positive integer. Then there is a constant C depending only on N , m and A such that for all f ∈ L+ A R∞ 0 0 S m,A ({x : S m ∗ f (x) ≥ t})dt ≤ C|||f |||A . 6.2. A space of quasicontinuous functions This subsection is devoted to generalize some results in [6] and in [55] relative to the Lp Lebesgue classes. From the previous Theorem, it is natural to seek when the quantity Z

∞ 0

0 Bm,A ({x : |ψ| ≥ t}) dt

(1)

defines a norm on a linear space of functions ψ on IR N . The answer is not known in general, but we establish that (1) is equivalent to a certain norm of ψ.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

252

Definition 6.1. For ψ a function on IRN , define for m > 0, Kψ and Λm,A (ψ) as N Kψ = f ∈ L + A : Gm ∗ f (x) ≥ |ψ(x)| , ∀x ∈ IR Λm,A (ψ) = inf {|||f |||A : f ∈ Kψ } .

Theorem 6.3. 1. Let A be any N -function and m ∈ IR + . The function Λm,A defines a norm on C0 = C0 (IRN ), and Z ∞ 1 0 Bm,A ({x : |ψ(x)| ≥ t}) dt Λm,A (ψ) ≤ 4 0 for all continuous functions ψ. 2. Let A be an N -function such that A and A∗ verify the ∆2 condition and m a positive integer. Then there is a constant C such that Z ∞ 1 0 Bm,A ({x : |ψ(x)| ≥ t}) dt ≤ CΛm,A (ψ) Λm,A (ψ) ≤ 4 0 for all continuous functions ψ.

0 We define a new Banach space, LA (Bm,A ), as the completion of D(IRN ) in the norm Λm,A .

Theorem 6.4. 1) Let A be an N -function such that A and A∗ verify the 0 ∆2 condition and m a positive integer. Then Lm,A ⊂ LA (Bm,A ) and any 0 continuous compactly supported function belongs on LA (Bm,A ). 2) Let A be an N -function such that A and A∗ verify the ∆ ∈ C is such that 2 condition and m a positive integer. If ψ R∞ 0 0 0 Bm,A ({x : |ψ(x)| ≥ t}) dt < ∞, then ψ ∈ LA (Bm,A ). Theorem 6.5. Let A be an N -function such that A and A∗ verify the ∆2 condition and m a positive integer. Then a function ψ on IR N belongs to 0 LA (Bm,A ) if and only if it is (m, A)-quasicontinuous, and Z ∞ 0 Bm,A ({x : |ψ(x)| ≥ t}) dt < ∞. 0

Corollary 6.1. Let A and m be as in the previous theorem. If ψ ∈ 0 LA (Bm,A ), and if ϕ is an (m, A)-quasicontinuous function such that |ϕ| ≤ |ψ| a.e., then 0 0 0 ), and kϕkLA (Bm,A ϕ ∈ LA (Bm,A ) ≤ kψkLA (Bm,A ). Moreover, if Lm,A is imbedded in a Banach space B such that k.kB is monotone in the sense that kukB ≤ ||v||B for all u and v such that |u(x)| ≤ 0 ). |v(x)| everywhere, then B contains LA (Bm,A

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

253 0 We describe now the dual space to LA (Bm,A ).

Theorem 6.6. Let A be an N -function such that A verifies the ∆2 condi0 tion and m > 0. Then the dual space LA (Bm,A )∗ can be identified with the N 0 space of all µ ∈ M(IR ) such that Gm ∗ |µ| ∈ LA∗ . If ψ ∈ LA (Bm,A ) and 0 ∗ 1 µ ∈ LA (Bm,A ) , then ψ ∈ L (|µ|), and the duality is given by hµ, ψi =

Z

ψdµ. RN

0 Moreover, the norm of µ in LA (Bm,A )∗ is ||Gm ∗ |µ| ||A∗ .

6.3. Maximal operators and capacity Let (θj )j be a sequence of convolution operators. Define the maximal operator J by J(f ) = supj |θj ∗ f |, where f is initially taken to be in the Schwarz class of rapidly decreasing C ∞ functions on IRN denoted by S = S(IRN ). An operator H : LA → LA is of strong type (A, A) if |||H(f )|||A ≤ C|||f |||A , ∀f ∈ LA , where C is a constant dependent only on A. For more details, see [93]. Definition 6.2. An operator H : LA → LA is of capacitary weak type (A, A) if 0 ({x : H(Im ∗ f )(x) ≥ t}) ≤ CA |||ft|||A , ∀f ∈ LA , ∀t > 0, Rm,A

where CA is a constant dependent only on N , m and A. H is of capacitary strong type (m, A) if ∀f ∈ LA , ,

R∞ 0

0 Rm,A ({x : H(Im ∗ f )(x) ≥ t}) dt ≤ C|||f |||A ,

where C is a constant dependent only on N , m and A. Theorem 6.7. 1. Let A be an N -function such that A and A∗ verify the ∆2 condition, α = α (A) and m is a positive integer such that m < N/α. If J is of strong type (A, A), then it is also of capacitary strong type (m, A). 2. Let A be any N -function. If J is of strong type (A, A) and 0 < m < N, then it is also of capacitary weak type (m, A).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

254

7. Maximal operators, Lebesgue points and quasicontinuity in strongly nonlinear potential theory We have shown that many maximal functions defined on some Orlicz spaces LA are bounded operators on LA if and only if they satisfy a capacitary weak inequality. We have shown also that (m, A)-quasievery x is a Lebesgue point for f in LA sense and we have given an (m, A)-quasicontinuous representative for f when LA is reflexive. See [35]. For i, j ∈ N, let θi,j be a complex valued function defined on IRN and such that θi,j ∈ LB for all N -functions B. Let the sequence (θj )j be such that (1) θi,j ∗ f → θj ∗ f in LB for all f ∈ LB (2) θj ∗ fn → θj ∗ f in LB if fn → f in LB . Define the maximal operator M M(f ) = supj |θj ∗ f | and assume that M(f ) is Lebesgue measurable on IR N . An operator H : LA → LA is of weak type (A,A) if ∀f ∈ LA , ∀t > 0, m({x : |H(f )(x)| > t}) ≤

A

1 Ct |||f |||A

where C is a constant dependent only on A, and m is the Lebesgue measure on RN . We say that H is of strong type (A,A) if ∀f ∈ LA , |||H(f )|||A ≤ C|||f |||A where C is a constant dependent only on A. Theorem 7.1. 1) Let A be an N -function and M the maximal operator defined by (7). Suppose M is of strong type (A,A). Then ∀f ∈ LA , ∀t > 0, Ck,A {x : M(k ∗ f )(x) > t}) ≤ A CA |||ft|||A . CA is the constant in the strong type. If we suppose in addition that A verifies the ∆2 condition, then there exists a constant C 0 dependent only on A, such that for all t > 0, |||f |||A 0 Ck,A ({x : M(k ∗ f )(x) > t}) ≤ C A . t

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

255

2) Let A be an N -function satisfying the ∆2 condition, and let M be the maximal operator defined by (7). Choose k = Gm with m > 0. Let C be a constant dependent only on A and such that for all t > 0 and all f ∈ L A , |||f |||A Ck,A ({x : M(Gm ∗ f )(x) > t}) ≤ CA . t Then M is of weak type (A, A). If in addition we suppose that A∗ verifies the ∆2 condition, then M is of strong type (A, A). 3) Let A be an N -function and let (ki )i be a sequence of positive integrable functions on IRN such that R a) R ki (x)dx → 1, as i → ∞ b) {|x|≥δ} ki (x)dx → 0, as i → ∞. 1 N . Then for any compact K in IR , limi→∞ Cki ,A (K) = A A−1 1 ( m(K) ) Theorem 7.2. Let A be an N -function such that A and A∗ satisfy the ∆2 condition and let α = α(A). Let m be a positive number and f = Gm ∗ g ∈ Lm,A , 0 < mα < N. Then (m, A)−quasievery x is a Lebesgue point for f in LA −sense, i.e. Z 1 e f (y)dy = f(x) exists, lim r→0 |B(x, r)| B(x,r)

and

lim r

r→0

−N α

|||fx |||A,B(x,r) = 0 ,

where fx is defined as fx (y) = f (y) − fe(x). Moreover, the convergence is uniform outside an open set of arbitrarily small (m, A)−capacity, fe is an (m, A)−quasicontinuous representative for f , and e = Gm ∗ g (m, A) − q.e. f(x)

8. Wolff inequality in strongly nonlinear potential theory and applications In this section we establish a Wolff type inequality for the strongly nonlinear potential theory. As applications, we give a relation between Bessel capacities and Hausdorff measure, and show that Riesz and Bessel capacities decrease under Lipschitz mapping in strongly nonlinear potential theory

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

256

for reflexive Orlicz spaces. This generalizes the similar result in the nonlinear case and a result in the strongly one when the Lipschitz mapping is an orthogonal projection. See [39]. 8.1. A Wolff type inequality The strongly nonlinear potential associated to Im and a positive measure µ is defined by VIµm ,A = Im ∗ a∗ (Im ∗ µ). The strongly nonlinear potential associated to Gm and µ is defined by VGµm ,A = Gm ∗ a∗ (Gm ∗ µ). µ We define Wm (x) = 1 ,A µ (x) = Wm ∞ ,A

R∞

R1

tm−1 a∗ (tm−N µ(B(x, t)))dt, and

0

tm−1 a∗ (tm−N µ(B(x, t)))dt.

0

Let K be a positive decreasing continuous function on ]0, ∞[. For x ∈ IR , x 6= 0, define K(x) = K(|x|). We suppose in addition that K satisfies K(r) ≤ LK(2r) for some L > 0, and all sufficiently small r > 0. This implies that L ≥ 1. Special cases of such K are Bessel and Riesz kernels. The strongly nonlinear potential associated to K and µ is defined by N

µ VK,A = K∗a∗ (K∗µ).

We set also WKµ1 ,A (x) = WKµ∞ ,A (x) =

R∞ 0

R1 0

K(t)tN −1 a∗ (K(t)µ(B(x, t)))dt, and

K(t)tN −1 a∗ (K(t)µ(B(x, t)))dt.

Lemma 8.1. Let A be an N -function such that A∗ verifies the ∆2 condition. Then there is a constant C such that for all positive measures µ, µ VK,A (x) ≥ CWKµ∞ ,A (x), ∀x ∈ IRN .

The reverse inequality is false in general, but we establish an inequality in term of integrals for Riesz and Bessel kernels. Hence we obtain the following Wolff type inequality. Theorem 8.1. Let A be an N -function such that A and A∗ satisfy the ∆2 condition and let µ be a positive Radon measure. Let 0 < m < N. There

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

257

are constants C and C 0 such that Z Z Z µ µ µ 0 Wm dµ C Wm dµ ≤ V dµ ≤ C Im ,A ∞ ,A ∞ ,A and C

Z

µ Wm dµ 1 ,A

≤

Z

VGµm ,A dµ

≤C

0

Z

µ Wm dµ. 1 ,A

8.2. Capacity and Hausdorff measure We have established the relation between Bessel capacities and Hausdorff measure and also a condition in term of Hausforff measure for a Bessel capacity of a set to be null. In this subsection we establish a converse deeper result that generalizes a theorem by V. P. Havin and V. G. Maz’ya [62, Theorem 7.1]. We begin by recalling some definitions about Hausforff measure. Let h : [0, +∞[→ IR be an increasing function satisfying h(0) = 0. For a subset X ⊂ IRN consider coverings of X by countable unions of (open or closed) balls {B(xi , ri }i≥1 with radii {ri }i≥1 . Let s be such that 0 < s ≤ ∞, P (s) (s) h(ri ), where the infimum and define a set function Λh by Λh (X) = inf i≤1

is taken over all such coverings with ri ≤ s for all i ≥ 1. (s) Because Λh (X) is a decreasing function of s, the Hausdorff measure of X with respect to the function h is defined by (s)

Λh (X) = lim Λh (X). s→0

(∞)

The Hausdorff content or the Hausdorff capacity is the set function Λh (∞) and we know that Λh (X) = 0 if and only if Λh (X) = 0. Recall that if m > J(A, N ), there exists a positive constant C = C(A, N, m) such that Bm,A (X) ≥ C for all set X such that X 6= ∅. Hence we must avoid this case when we work with sets with null capacity. Theorem 8.2. Let A be an N -function such that A and A∗ satisfy the ∆2 condition and X be a subset of IRN . Let 0 < m ≤ J(A, N ), and R1 let h be an increasing function on [0, +∞[ satisfying h(0) = 0, and 0 tm−1 a∗ (tm−N h(t))dt < ∞. Then Λh (X) = 0 if Bm,A (X) = 0.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

258

8.3. Lipschitz mappings and capacities We know that Bessel (and also Riesz) capacities in reflexive Orlicz spaces are non increasing under orthogonal projection of sets. Here we generalize this result to Lipschitz maps. Theorem 8.3. Let A be an N -function such that A and A∗ satisfy the ∆2 condition and let 0 < m < N. Assume that E ⊂ R N and that Φ : E → IRN is a Lipschitz mapping, i.e. there is a constant L such that Φ satisfies |Φ(x) − Φ(y)| ≤ L |x − y| for all x, y ∈ E. Then there is a constant C depending only on L, m, N and A, such that Bm,A (Φ(E)) ≤ CBm,A (E) and Rm,A (Φ(E)) ≤ CRm,A (E). 9. On the A-Laplacian We establish, for Orlicz spaces LA (IRN ) such that A satisfies the ∆2 condition, theR non resolubility of the A-Laplacian equation ∆A u + h = 0 on IRN , and h 6= 0, if IRN is A-parabolic. For a large class of Orlicz spaces including Lebesgue spaces Lp (p > 1), we prove also that the same equation, with any bounded measurable function h with compact support, has a solution with gradient in LA (IRN ) if IRN is A-hyperbolic. See [37]. Definition 9.1. Let A be an N -function and K a compact set in IR N . The A-capacity of K is defined by ΓA (K) = inf |||∇u|||A : u ∈ C0∞ (IRN ), u = 1 in a neighborhood of K .

The space IRN is said to be A-parabolic if ΓA (K) = 0 for all compact subsets K ⊂ IRN and A-hyperbolic otherwise.

The A-Dirichlet space L1A (IRN ) is the space of functions u ∈ 1 (IRN ) (i.e. u is locally in W 1 LA (IRN )) admitting a weak gradient WA,loc such that |||∇u|||A < ∞. Let A be any N -function and let a be its derivative. The A-Laplacian a(|∇f |) N of a function f on IR is defined by ∆A f = div |∇f | .∇f . 1 (IRN ) is said to be a weak solution to the equation A function u ∈ WA,loc

∆A u + h = 0

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

259

if for all ϕ ∈ C01 (IRN ), we have Z Z a(|∇u|) h .∇u, ∇ϕi dλ = hϕ dλ. |∇u|

Let D ⊂ IRN be a non empty bounded domain. The Banach space 1 EA (D) is the space of functions u ∈ WA,loc (IRN ) such that D

|||∇u|||A := |||u|||A,D + |||∇u|||A < ∞. 0 We denote by EA (D) the closure of C01 (IRN ) in EA (D).

9.1. A non resolvability result Theorem 9.1. Let A be an N -function satisfying the ∆2 condition. SupR pose that IRN is A-parabolic and let h ∈ L1 (IRN ) be such that h dλ 6= 0. Then the equation ∆A u + h = 0

(2)

has no weak solution on L1A (IRN ). Corollary 9.1. Let LA (IRN ) be a Rreflexive Orlicz space such that α∗ ≤ N N h dλ 6= 0. Then the equation (2) has N −1 . Let h ∈ L1 (IR ) be such that no weak solution on L1A (IRN ). Remark 9.1. When A(t) = p−1 tp , LA = Lp is the usual Lebesgue space p and α∗ = p−1 . Hence the condition α∗ ≤ NN−1 is exactly the condition N ≤ p. Thus our result recovers the one in [58]. 9.2. A resolvability result In this subsection we resolve the equation ∆A u + h = 0 under some assumptions on the N -function A, and on the function h. We begin by the following type Poincar´e inequalities for Orlicz-Sobolev functions. See [36]. Theorem 9.2. 1) Let A be an N -function such that A and A∗ satisfy the ∆2 condition. Let E be any measurable set in IR N , such that 0 < λ(E) < ∞. Then there exists a positive constant C such that |||u − uE |||A,E ≤ C |||∇u|||A,E , R 1 1 for all u ∈ WA,loc (IRN ), where uE = λ(E) u dλ is the mean value of u on E

E. 2) Let A be an N -function such that A and A∗ satisfy the ∆2 condition.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

260

Let E be any measurable set in IRN , such that 0 < λ(E) < ∞. Then there exists a positive constant C such that Z |u − uE | dλ ≤ C |||∇u|||A,E , E

1 for all u ∈ WA,loc (IRN ). 3) Let A be an N -function such that A and A∗ satisfy the ∆2 condition. Suppose that IRN is A-hyperbolic. Let E be any non empty bounded domain 0 in IRN . Then there exists a positive constant C such that for all u ∈ EA (E) Z |u| dλ ≤ C |||∇u|||A . E

R Recall that for all f ∈ Ln that |||f ||| > 1, we have Ao◦ f dλ > A such A R log A◦f dλ |||f |||A . We set s(A) = inf log|||f ||| − 1, f ∈ LA , |||f |||A > 1 . Hence A

s(A) ≥ 0. One can consult [37] for some examples of N -functions satisfying s(A) > 0. Now we are ready to solve the A-Laplace equation.

Theorem 9.3. Let LA be a reflexive Orlicz space such that s(A) > 0. Let h ∈ L∞ (IRN ) have compact support. Then the equation ∆A u + h = 0 has a weak solution u ∈ L1A (IRN ) if IRN is A-hyperbolic. 0 Remark 9.2. 1) We have in fact solved the equation in the space EA (D) ⊂ 1 N LA (IR ). 2) When A(t) = p−1 tp , p > 1, LA = Lp is the usual Lebesgue space, we have s(A) = p − 1 > 0. Thus we recover the result in [94], when the manifold M is IRN .

Corollary 9.2. Let LA (IRN ) be a reflexive Orlicz space such that s(A) > 0 and α < N . Suppose that h ∈ L∞ (IRN ) has compact support. Then the equation ∆A u + h = 0 has a weak solution u ∈ L1A (IRN ). References 1. D. R. Adams, Traces of potentials arising from translation invariant operators, Ann. Scuola Norm. Sup. Pisa Cl. Sci. 25 (1971) 203-217. 2. D. R. Adams, Maximal operators and capacity, Proc. Amer. Math. Soc., 34 (1972) 152-156. 3. D. R. Adams, Traces of potentials. II, Indiana Univ. Math. J., 22 (1973) 907-918. 4. D. R. Adams, A note on Riesz potentials, Duke Math. J., 42 (1975) 765-778.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

261

5. D. R. Adams, On the existence of strong type estimates in IR N , Ark. Mat., 14 (1976) 125-140. 6. D. R. Adams, Sets and functions of finite Lp -capacity, Indiana Univ. Math. J., 27 (1978a) 611-627. 7. D. R. Adams, Quasi-additivity and sets of finite Lp -capacity, Pacific Journal Math., 79 (1978) 283-291. 8. D. R. Adams, Capacity and the obstacle problem, Appl. Math. Optim., 8 (1981) 39-57. 9. D. R. Adams, Lectures on Lp -Potential Theory, Department of Mathematics, University of Ume˚ a, 1981. 10. D. R. Adams, The exceptional sets associated with the Besov spaces, in Linear and Complex Analysis Problem Book. 199 Research Problems (V. P. Havin, S. V. Hruˇsˇc¨ev, N. K. Nikol’skii, eds.), Lecture Notes in Math. 1043, 515518, Springer, Berlin Heidelberg, 1984. Also in Linear and Complex Analysis Problem Book 3, Part II (V. P. Havin, N. K. Nikol’skii, eds.) Lecture Notes in Math. 1574, 169-172, Springer, Berlin Heidelberg, 1994. 11. D. R. Adams, Weighted nonlinear potential theory, Trans. Amer. Math. Soc. 287 (1986) 73-94. 12. D. R. Adams, A note on the Choquet integrals with respect to Hausdorff capacity, in Function Spaces and Applications, Proc. Lund 1986 (M. Cwikel, J. Peetre, Y. Sagher, H. Wallin, eds.) Lecture Notes in Math. 1302, 115-124, Springer, Berlin Heidelberg, 1988. 13. D. R. Adams, A sharp inequality of J. Moser for higher order derivatives, Ann. of Math., 128 (1988) 385-398. 14. D. R. Adams and A. Heard, The necessity of Wiener test for some semi-linear elliptic equations, Indiana Univ. Math. J., 41 (1992) 109-124. 15. D. R. Adams and L. I. Hedberg, Inclusion relation among fine topologies in nonlinear potential theory, Indiana Univ. Math. J., 33 (1984) 117-126. 16. D. R. Adams, L. I. Hedberg, Function spaces and potential theory. SpringerVerlag, Berlin, New York 1996. 17. D. R. Adams and J. L. Lewis, Fine and quasi connectedness in nonlinear potential theory, Ann. Inst. Fourier (Grenoble) 35:1 (1985) 57-73. 18. D. R. Adams and N. G. Meyers, Thinness and Wiener criteria for nonlinear potentials, Indiana Univ. Math. J., 22 (1972) 169-197. 19. D. R. Adams and N. G. Meyers, Bessel potentials. Inclusion relations among classes of exceptional sets, Indiana Univ. Math. J., 22 (1973) 873-905. 20. D. R. Adams and M. Pierre, Capacitary strong type estimates in semi-linear problems, Ann. Inst. Fourier (Grenoble) 41 (1991) 117-135. 21. D. R. Adams and J. C. Polking, The equivalence of two definitions of capacity, Proc. Amer. Math. Soc., 37 (1973) 529-534. 22. R. Adams, Sobolev spaces, Acad. Press, New York, London, Torento, (1975). 23. R. Adams, On the Orlicz-Sobolev imbedding theorem, J. Func. Analysis, 24 (1977) 241-257. 24. H. Aikawa, Tangential boundary behavior of Green potentials and contractive properties of Lp -capacities, Tokyo Math. J. 9 (1986) 223-245. 25. H. Aikawa, Comparison of Lp -capacity and Hausdorff measure, Complex

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

262

Variables, 15 (1990)223-232. 26. H. Aikawa, Quasiadditivity of Riesz capacity, Math. Scand., 69 (1991) 15-30. 27. H. Aikawa, Quasiadditivity of capacity and minimal thinness, Ann. Acad. Sci. Fenn. Ser. A. I. Math. 18 (1993) 65-75. 28. N. A¨ıssaoui, Note sur la capacitabilit´e dans les espaces d’Orlicz, Ann. Sci. Math. Qu´ebec. 19 (1995) 107-113. 29. N. A¨ıssaoui, On the continuity of Bessel potentials in Orlicz spaces, Collect. Math. 47, 1 (1996) 77-89. 30. N. A¨ıssaoui, Instability of capacity in Orlicz spaces, Potential analysis, 6 (1997) 327-346. 31. N. A¨ıssaoui, Bessel potentials in Orlicz spaces, Revista Matem´ atica de la Universidad Complutense de Madrid, Vol. 10, No .1 (1997) 55-79. 32. N. A¨ıssaoui, Some developments of strongly nonlinear potential theory, Libertas Mathematica, Vol.XIX (1999) 155-170. 33. N. A¨ıssaoui, On the continuity of strongly nonlinear potential, Math. Slovaca, 51 (2001), no 2, 215-228. 34. N. A¨ıssaoui, Capacitary type estimates in strongly nonlinear potential theory and applications, Revista Matem´ atica de la Universidad Complutense de Madrid, Vol. XIV, num.2 (2001) 347-370. 35. N. A¨ıssaoui, Maximal operators, Lebesgue points and quasicontinuity in strongly nonlinear potential theory, Acta Math. Univ. Comenianae, Vol LXXI, 1 (2002) 35-50. 36. N. A¨ıssaoui, Strongly nonlinear potential theory on metric spaces, Abstr. Appl. Anal., 7, no. 7, (2002) 357-374. 37. N. A¨ıssaoui, On the A-Laplacian, Abstr. Appl. Anal., 2003:13 (2003) 743755. 38. N. A¨ıssaoui, Weighted strongly nonlinear potential theory, Houston Journal of Mathematics, 29, no. 1, (2003) 207-230. 39. N. A¨ıssaoui, Wolff inequality in strongly nonlinear potential theory and applications, African Diaspora Journal of Mathematics, Vol1, Number 1, (2004) 69-83. 40. N. A¨ıssaoui, Another extension of Orlicz-Sobolev spaces to metric spaces, Abstr. Appl. Anal., 2004:1, (2004) 1-26. 41. N. A¨ıssaoui, Orlicz-Sobolev spaces with zero boundary values on metric spaces, Southwest Journal of Pure and Applied Mathematics, 1, (2004) 10-32. 42. N. A¨ıssaoui, A. Benkirane, Capacit´es dans les espaces d’Orlicz, Ann. Sci. Math. Qu´ebec. 18 (1994) no 1, 1-23. 43. N. A¨ıssaoui, A. Benkirane, Potentiel non lin´eaire dans les espaces d’Orlicz, Ann. Sci. Math. Qu´ebec. 18 (1994) no 2, 105-118. 44. N. Aronszajn, F. Mulla, P. Szeptycki, On spaces of potentials connected with Lp classes, Ann. Inst. Fourier (Grenoble) 13 (1963) 211-306. 45. N. Aronszajn, K.T Smith, Theory of Bessel potentials, Ann. Inst. Fourier (Grenoble) 11 (1961) 383-475. 46. A. Benkirane, Potentiel de Riesz et probl`emes elliptiques dans les espaces d’Orlicz, Th`ese de Doctorat, U. L. B. (1988). 47. A. Benkirane and J. P. Gossez, An approximation theorem in higher order

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

263

Orlicz-Sobolev spaces and applications, Studia Math., 92 (1989) 231-255. 48. C. Fernstr¨ om, On the instability of capacity, Ark. Mat., 15 (1977) 241-252. 49. B. Fugled, On the theory of potentials in locally compact spaces, Acta Math., 103 (1960) 139-215. 50. B. Fugled, On generalized potentials of functions in the Lebesgue classes, Mat. Scand., 8 (1960) 287-304. 51. B. Fugled, Le th´eor`eme du minimax et la th´eorie fine du potentiel, Ann. Inst. Fourier (Grenoble) 15:1 (1965) 65-88. 52. B. Fugled, Quasi topology and fine topology, S´eminaire de Th´eorie du Potentiel (Brelot-Choquet-Deny) 10 (1965-1966) no. 12. 53. B. Fugled, Applications du th´eor`eme minimax a ` l’´etude de diverses capacit´es, C. R. Acad. Sci. Paris S´er. A-B 266 (1968) 921-923. 54. B. Fugled, The quasi topology associated with a countably subadditive set function, Ann. Inst. Fourier (Grenoble) 21:1 (1971) 123–169. 55. B. Fugled, Connexion en topologie fine et balayage des mesures, Ann. Inst. Fourier (Grenoble) 21:3 (1971) 227–244. 56. B. Fugled, Finely harmonic functions, Lecture Notes in Math. 289, Springer, Berlin Heidelberg, 1972. 57. B. Fugled, A simple proof that certain capacities decrease under contraction, Hiroshima Math. J. 19 (1989) 567-573. 58. V. Gol’dshtein, M. Troyanov, Sur la non r´esolubilit´e du p-Laplacien sur IR n , C. R. Acad. Sci. Paris 326 (1998), 1185-1187. 59. K. Hansson, Imbedding theorems of Sobolev type in potential theory, Math. Scand., 45 (1979) 77-102. 60. V. P. Havin, Non-linear potential theory, Russian Math. Surveys 27:6 (1972) 71-148. 61. V. P. Havin, Use of (p,l)-capacity in problems of the theory of exceptional sets, Math. USSR-Sb., 19 (1973) 547-580. 62. V. P. Havin and V. G. Maz’ya, Non-linear potential theory, Russian Math.Surveys 27:6 (1972) 71-148. 63. L. I. Hedberg, Non-linear potentials and approximation in the mean by analytic functions, Math. Z. 129 (1972) 299-319. 64. L. I. Hedberg, On certain convolution inequalities, Proc. Amer. Math. Soc., 36 (1972) 505-510. 65. L. I. Hedberg, Bounded point evaluation and capacity, J. Funct. Anal., 10 (1972) 269-280. 66. L. I. Hedberg, Approximation in the mean by solutions of elleptic equations, Duke Math. J., 40 (1973) 9-16. 67. L. I. Hedberg, removable singularities and condenser capacities, Ark. Mat., 12 (1974) 181-201. 68. L. I. Hedberg, Two approximation problems in function spaces, Ark. Mat., 16 (1978) 51-81. 69. L. I. Hedberg, Nonlinear potential theory, in Potential Theory, Proc. Int. Conf. Potential Theory, Nagoya 1990 (M. kishi, ed.) 43-54, de Gruyter, Berlin New York, 1991. 70. L. I. Hedberg and Th. H. Wolff, Thin sets in nonlinear potential theory, Ann.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

264

Inst. Fourier (Grenoble) 33:4 (1983) 161-187. 71. B. Jawerth, C. P´erez, G. Welland, The positive cone in Triebel-Lizorkin spaces and the relations among potential and maximal operators. Contemporary Mathematics, Proc. Conf. Boca Raton 1988 (M. Milman and T. Schonbek, Eds.) 107, 71-971. Amer. Math. Soc., Providence, R. I, 1989. 72. M. A. Krasnosel’skii, Y. B. Rutickii, Convex functions and Orlicz spaces, P. Noordhoff, Groningen (The Netherlands), 1961. 73. A. Kufner, O. Hohn, S. Fuˇc´ık, Function spaces. Academia, Prague, 1977. 74. W. A. J. Luxemburg, Banach function spaces. Thesis, Delft University, 1955. 75. V. G. Maz’ya, On the theory of n-dimensional Schrodinger operator, Izv. Akad. Nauk SSSR. Ser.Mat. 28 (1964) 1142-1172. 76. V. G. Maz’ya, Polyharmonic capacity in the theory of the first boundary value problem, Sibirsk. Mat. Zh. 6 (1965) 127-148. 77. V. G. Maz’ya, On (p,l)-capacity, imbedding theorems and the spectrum of a self adjoint operators, Mat. USSR-Izv. 7 (1973) 375-387. 78. V. G. Maz’ya, Behaviour of solutions to the Dirichlet problem for the biharmonicoperator at a boundary point, Lecture Notes in Math. 703 , 250-262, Springer, Berlin Heidelberg, 1979. 79. V. G. Maz’ya, Sobolev spaces, Springer-Verlag, Berlin-New York, 1985. 80. N. G. Meyers, A theory of capacities for potentials of functions in Lebesgue classes, Math. Scand., 26 (1970) 255-292. 81. N. G. Meyers, Continuity of Bessel potentials, Isr. J. Math. 11 (1972) 271283. 82. N. G. Meyers, Taylor expansion of Bessel potentials, Indiana Univ. Math. J., 23 (1974) 1041-1049. 83. N. G. Meyers, Continuity properties of potentials, Duke Math. J., 42 (1975) 157-166. 84. N. G. Meyers, Integral inequalities of Poincar´e and Wirtinger type, Arch. Rat. Mech. Anal., 68 (1978) 113-120. 85. B. Muckenhoupt and R.L. Wheeden, Weighted norm inequalities for fractional integrals, Trans. Amer. Math. Soc., 192 (1974) 261-274. 86. Yu. G. Reshetnyak, On the concept of capacity in the theory of functions with generalized derivatives, Sibirsk Mat. Z., 10 (1969) 1109-1138. 87. Yu. G. Reshetnyak, On the boundary behavior of functions with generalized derivatives, Sibirsk. Mat. Zh. 13 (1972), 411-419. English translation: Sibirian Math. J. 13 (1972), 285-290. 88. M. M. Rao, Z. D. Ren, Theory of Orlicz spaces, Dekker Inc, New York, 1991. 89. J. Serrin, Local behavior of solutions of quasi-linear equations, Acta Math., 111 (1964) 247-302. 90. J. Serrin, Removable singularities of solutions of elliptic equations, Arch. Rational Mech. Anal., 17 (1964) 65-78. 91. T. Sj¨ odin, Bessel potentials and extension of continuous functions on compact sets, Ark. Mat., 13 (1975) 263-271. 92. E. M. Stein, Singular integrals and differentiability properties of functions, Princeton Univ. Press, Princeton, New Jersey 1970. 93. A. Torchinsky, Interpolation of operators and Orlicz classes, Studia Math.,

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

265

59 (1976) 177-207. 94. M. Troyanov, Solving the p-Laplacian on manifolds, Proc. Amer. Math. Soc. 128 (2000) no 2, 541-545. 95. B. O. Turesson, Nonlinear Potential Theory and Weighted Sobolev Spaces. Lecture Notes in Mathematics 1736, Springer-Verlag, Berlin, Heidelberg, New York, 2000. 96. W. P. Ziemer, Weakly differentiable functions, Graduate Texts in Mathematics, Vol. 120, Springer-Verlag, New York, 1989.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

266

Back on stochastic model for sandpile Noureddine Igbida LAMFA, UMR 6140, Universit´ e de Picardie Jules Verne 33 rue Saint Leu, 80038 Amiens, France E-mail: [email protected] Our aim in this note is to give a simplified proof of the convergence of EvansRezakhanlou Stochastic Model to the evolution surfaces model of sandpile. Keywords: Evans-Rezakhanlou Stochastic Model; Convergence.

1. Introduction The stochastic model for sandpile was introduced by Evans and Rezakhanlou in6 as a variant ofphysical models for sandpile. It corresponds to a Markov process X(t) defined by an infinitesimal generator describt≥0

ing the evolution of stack of unit cubes resting on the plane when new cubes are being added to the pile, by being placed either upon a heretofore unoccupied square in the plane or else upon the top of a current column.

At each time t > 0, the configuration X(t) needs to be stable which means that the heights of any two adjacent columns of cubes can differ by at most

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

267

one. So adding new cubes on existing pile, we ordain two possibilities to each cube: • if the configuration is stable, then the cube remains in place • otherwise, the cube has several downhill ”staircases” along which it can move, and the cube will randomly select among the allowable downhill paths. If we consider the case where the cubes can be also taken away, then a third possibility may be ordain • the cube has several ”upward staircases” along which it can move, and the cube will randomly select among the allowable upward paths. Assuming that we continuously add cubes at random locations on a starting empty stack, Evans and Rezakhanlou studies the limit when one rascals in both space and time, so as to consider growing piles of more and more smaller and smaller cubes ? They prove that the macroscopic limit is rather simple and very connected Prigozhin model for sandpile (cf.8 and2 ). Since the work of Prigohzin (cf.8 see also2 ), this has been well known (see also,36 and the references therein) that the evolution of the surface of the sandpile when the angle of stability is equal to π/4 can be described by the following evolution problem   ∂t u + ∂IIK (u) 3 f (1)  u ˆ(0) = 0, n o where K = z ∈ W 1,∞ (IR2 ) ∩ L2 (IR2 ) ; |∇z| ≤ 1 , ∂IIK denotes the sub-

differential operator of the indicator function (cf.4 ) of K and f describes a source term . Solution u is the height of the surface that grows up (resp. grows down) under sand addition (resp. sand removal) by a source called f . This is a critical slope model obtained by using the continuity equation and the gradient constraint |∇u| ≤ 1 a.e. in IR2 (see8 and2 for more details). Existence, uniqueness and numerical approximation the solution are well known by now for this model (cf.,8 ,25 and7 ). More precisely, for any f ∈ BVloc (0, T ; L2 (IR2 )), we know that that (1) has a unique solution u in the 1,∞ sense that u ∈ Wloc (0, ∞; L2 (IR2 )), u(0) = 0 and, for any t ≥ 0, u(., t) ∈ K and Z f (t, x) − ∂t u(t, x) u(t, x) − ξ(x) ≥ 0 for any ξ ∈ K. (2) IR2

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

268

To give the connection between the two models, let N be a large integer. Assume the cubes are of side length O(N −1 ) and cubes newly and randomly added at rate O(N −1 ) are continuously falling downhill. For the description of the evolving of cubes, the authors of 6 introduce a probabilistic lattice model. They thus consider a Markov process for the height X(t, i), defined t i for times t ≥ 0 and sites i ∈ ZZ 2 . Resacaled source terms f + and , N N t i control the rate new cubes are added to the pile or removed f− , N N from it. Then, they proved that 2 i hZ 1 (3) IE X N t, [N x] − u(t, x) → 0 as N → ∞, IR2 N

where u is the unique solution of (1). As shown in,6 the key estimates for the proof (3) are some kind of microscopic version of (2) (see (10)). To prove these estimates the authors of 6 prove some elementary intermediate estimates evolving various types of sets. Our aim in this note, is to improve these key estimates directly by using some simple arguments. In the next section, we recall the stochastic model of Evans and Rezakhanlou. In Section 3, we prove the key estimates (10). Then, as in,6 we introduce the discrete evolution problem associated with (1). Then, we give the proof of (3) by using some results of.1 2. The stochastic model for sandpile problem (cf.6 ) To describe the stochastic process for sandpile, we consider the lattice ZZ n . We equipped ZZ n with an Euclidean norm and we say that i, j ∈ ZZ n are adjacent, written i ∼ j, provided |i − j| ≤ 1. Without loose of generality, we restrict ourself to the cases n = 2, and we write i = (i1 , i2 ) to denote a typical site in ZZ 2 . Then, we introduce the Hilbert space n o X H := l2 (ZZ 2 ) = X : ZZ 2 → IR ; kXk := X(i)2 < ∞ . i

2

A (stable) configuration is a mapping X : ZZ → ZZ such that   |X(i) − X(j)| ≤ 1 if i ∼ j 

and X has bounded support.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

269

The state space is n o S := X : ZZ 2 → ZZ ; X is a configuration ,

and the set of stable configuration is n o ˆ := X ∈ H ; |X(i) − X(j)| ≤ 1 if i ∼ j . K

Let fˆ : (0, ∞) × ZZ 2 → ZZ be a function, such that f + (resp. f − ) is controlling the rate new cubes are added (resp. removed) to the pile. The source of cubes fˆ generates a stochastic process (X(t), t ≥ 0) in the state space S. It is clear that the probability that X(t) be situated (at time t) in a given set Γ of E, under the condition that the movement of the system up to time s (s < t) is completely known, depends only on the state of the system at time s. In other words (X(t), t ≥ 0) is a Markov process. To study this process, we need to know its infinitesimal generator A, or more precisely AF (X(t)) for any F ∈ B(S), the set of bounded functions. To this aim, let us consider p+ (i, j, ξ) the probability that a cube placed on a given configuration ξ ∈ S at the position i will end up at j after it has fallen downward over the stack with eight ξ. Likewise, let p− (i, j, ξ) be the probability that the removal of a cube from the pile at i will result in a removal at site j, after the cubes along a staircase each shifts downwards to fill in the gap created at site i. So, for any i, j ∈ ZZ 2 we have X 0 ≤ p± (i, j, ξ) ≤ 1 and p± (i, j, ξ) = 1. i∈Z Z2

Thanks to,6 we consider c± given by X c± (j, X, t) = p± (i, j, X(t)) fˆ± (t, i), i∈Z Z2

for any (t, j) ∈ ZZ 2 × [0, ∞).

(4) The parameter c (j, X, τ ) (resp. c (j, X, τ )) is highly nonlocal factor and records the rate, at time τ, new cubes come to rest at the site j after falling downhill (resp. the rate at which cubes are removed from the site j, at time τ ). In particular, we have +

−

X

j∈Z Z2

c(j, X, t) =

X

f (t, i).

i∈Z Z2

Thanks again to,6 the infinitesimal generator A of the Markov process

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

270

(X(t), t ≥ 0), is given by X A F (X(t)) = c+ (j, X, t)(F (X(t) + δj ) − F (X(t))) j∈Z ZN

−

X

j∈Z ZN

−

c (j, X, t)(F (X(t) − δj ) − F (X(t)))

(5)

for any t ≥ 0,

where, for any j ∈ ZZ 2 , δj : ZZ 2 → IN is given by   1 if i = j δj (i) =  0 otherwise. In particular, we have

AF (X(t)) = (Lt F )(X(t))

for any (t, F ) ∈ (0, ∞) × B(S),

(6)

where Lt : F ∈ B(S) → Lt F ∈ L(S, IR), for any t ≥ 0, is the time dependent operator defined by (Lt F )(ξ) :=

X

c+ (j, ξ, t)(F (ξ+δj )−F (ξ))−

j∈Z ZN

X

c− (j, ξ, t)(F (ξ−δj )−F (ξ)).

j∈Z ZN

(7)

3. Main results Thanks to,6 we know that the connection between the stochastic model and (1) is given through the following nonlinear dynamic in l 2 (ZZ 2 ) :  ˆ + IIKˆ (ˆ u) 3 fˆ for t ≥ 0  ∂t u (8)  u ˆ(0) = 0,

ˆ is closed where ∂IIKˆ denotes the sub-differential of IIKˆ in l2 (ZZ 2 ). Since K 2 4 ˆ ˆ and convex with S ⊆ K, then (cf. ) for a given f ∈ BV (0, T ; l (ZZ 2 )), the 1,∞ evolution (8) has a unique solution u ˆ ∈ Wloc (0, ∞; H) such that u ˆ(0) = 0 ˆ and and, for any t ≥ 0, u ˆ(., t) ∈ K X ˆ ˆ fˆ(t, i) − ∂t u ˆ(t, i) u ˆ(t, i) − ξ(i) ≥ 0 for any ξˆ ∈ K. i

We have

Theorem 3.1. Assume that fˆ ∈ BV (0, T ; l2(ZZ 2 ). Let u ˆ be the solution of (8) and (X(t), t ≥ 0) be the stochastic process generated by fˆ. Then, we

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

271

have IE

"

X

i∈Z Z2

(X(t, i) − u ˆ(t, i))

2

#

≤

Z

t

X

0 j∈Z Z2

ˆ s) ds, f(j,

for any t ≥ 0. (9)

Thanks to,6 recall that the proof of Theorem 3.1 is based on the following estimates.

Lemma 3.1. Under the assumptions of Theorem 3.1, for any w ∈ S, we have X X ± c± (j, X, t)(X(t, j)−w(j)) ≤ ± fˆ(t, i)(X(t, i)−w(i)) ∀w ∈ S. j∈Z Z2

i∈Z Z2

(10)

The estimates (10) are some kind of microscopic version of (2). To prove (10) the authors of 6 prove some elementary intermediate estimates evolving various types of sets. In these notes, we use essentially the following remark and give a direct and short proof of this result. Then, the proof of Theorem 3.1 follows more or less the same step of.6 Indeed, Lemma 3.1 gives the connection between X(t) and (8). Then, we transform (8) into an evolution problem in L2 (IR2 ) governed by a sub-differential operator of the indicator function of the set of rescaled configurations (see (14)). This transformation allows us to give the connection between (1), (15) and the stochastic model. Remark 1. (1) For a given ξ ∈ S, if p+ (i, j, ξ) > 0, then there exists at least one staircase i0 = i ∼ i1 ∼ . . . ∼ im = j, such that ξ(ip ) = ξ(ip+1 ) + 1 for any p = 0, 1, . . . , m − 1. Let us denote this staircase by C + (i, j) ; i.e. C + (i, j) = [i0 , i1 , . . . , im−1 , j], ˜ the adjacent side to k such and, for any k ∈ C + (i, j), we denote by k, ˜ that u(k) = u(k) + 1. It is clear that C(i, j) may not be unique, so (it is not essential) to simplify the presentation, let us consider the application: C + : ZZ 2 × ZZ 2 → C + (i, j)a staircase between i and j.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

272

(2) For a given ξ ∈ S, if p− (i, j, ξ) > 0, then there exists at least one upward staircase i0 = i ∼ i1 ∼ ... ∼ im = j, such that ξ(ip ) = ξ(ip+1 ) + 1 for any p = 0, 1, . . . , m − 1. Let us denote this staircase by C − (i, j) ; i.e. C − (i, j) = [i0 , i1 , ..., im−1 , j],

˜ the adjacent side to k such that and, for any k ∈ C(i, j), we denote by k, ˜ − 1. It is clear that C(i, j) may not be unique, so (it is not u(k) = u(k) essential) to simplify the presentation, let us consider the application : C − : ZZ 2 × ZZ 2 → C − (i, j) a staircase between i and j. Proof of Lemma 3.1: Thanks to (4), we have P + j∈Z Z 2 c (j, X, t) (X(t, j) − w(j)) P

=

= +

j,i∈Z Z2

p+ (i, j, X(t)) fˆ+ (t, i) (X(t, j) − w(j))

j,i∈Z Z2

p+ (i, j, X(t)) fˆ+ (t, i) (X(t, i) − w(i))

j,i∈Z Z2

p+ (i, j, X(t)) fˆ+ (t, i) (w(i) − w(j)) − (X(t, i) − X(t, j))

P

P

= I1 + I2 X p+ (i, j, X(t)) = 1, for any (t, i) ∈ ZZ 2 × (0, ∞), then it is clear Since j∈Z Z2

that

I1 =

X

i∈Z Z2

fˆ+ (t, i) (X(t, i) − w(i)).

Let us prove that I2 ≤ 0. Thanks to Remark 1, we have I2 =

X

p+ (i, j, X(t)) fˆ+ (t, i)

j,i∈Z Z2

≤

X

j,i∈Z Z2

X

k∈C + (i,j)

p+ (i, j, X(t)) fˆ+ (t, i)

X

k∈C + (i,j)

˜ ˜ t)) (w(k)−w(k))−(X(t, k)−X(k, ˜ −1 (w(k) − w(k))

≤ 0,

˜ ≤ 1 (since w ∈ K and k ∼ k). ˜ where we used the fact that |w(k) − w(k)| The proof of X X c− (j, X, t) (X(t, j) − w(j)) ≥ fˆ− (t, i) (X(t, i) − w(i)), j∈Z Z2

i∈Z Z2

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

273

follows in the same way, we let the details to the reader. To simplify the presentation of the proof of Theorem 3.1, we divide the proof into two steps that we present in the following Lemmas. Lemma 3.2. Under the assumptions of Theorem 3.1, we have (1) For any w ∈ S and t ≥ 0, we have 2 X 1 X 1 X f (t, j). X(t, i)−w(i) ≤ fˆ(t, j)(X(t, j)−w(j))+ Lt 2 2 2 2 2 j∈Z Z

i∈Z Z

j∈Z Z

(2) For any w ∈ W 1,∞ (0, T ; H) such that w(t) ∈ S, for any t ∈ [0, T ), we have RthP P ∂w 1 2 i∈Z Z 2 (X(t, i) − w(t, i)) ≤ 0 i∈Z Z 2 ∂s (i, s) (w(i, s) − X(i, s)) 2 i P f (j, s) (X(j, s) − w(j, s)) + 12 j∈ZZ 2 f (j, s) + M(t) where M(t) is a martingale satisfying +

P

j∈Z Z2

t≥0

IE(M(t)) = 0

for any t ≥ 0.

Proof: (1) Let us denote I=

2 1 X Lt X(t, i) − w(i) 2 2 i∈Z Z

By definition of L, we have X 2 X 2 1 X c(j, X, t) Tj (X(t))(i)−w(i) − X(t, i)−w(i) . I= 2 2 2 2 j∈Z Z

i∈Z Z

i∈Z Z

So, I=

X 1 X c(j, X, t) (Tj (X)(i) − X(i))(Tj (X)(i) + X(i) − 2 u(i)) 2 2 j∈Z Z

i∈Z Z

1 X c(j, X, t) (2 X(j) + 1 − 2 u(j)) = 2 2 j∈Z Z

=

X

j∈Z Z2

c(j, X, t) (X(j) − u(j)) +

1 X c(j, X, t). 2 2 j∈Z Z

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

274

Thanks to (4), we deduce that X 1 X I= f (t, j), c(j, X, t) (X(j) − u(j)) + 2 2 2 j∈Z Z

j∈Z Z

and, using Lemma 3.1, the first step of the lemma follows. (2) As in,6 we use the following stochastic integral equation : for any F : S ×(0, ∞) → IR Lipchitz continuous in t and F (X(., 0), 0) = 0, we have Z t ∂F F (X(., t), t) = + Ls F (X(., s)) + M(t). (11) ∂s 0 Let F be given by 2 1 X ξ(i) − w(t, i) , F (ξ, t) = 2 2 i∈Z Z

for any (ξ, t) ∈ S × (0, T ).

Then, X ∂w ∂F (ξ, s) = − (i, s) ξ(i)−w(i, s) , ∂s ∂s 2 i∈Z Z

for any (ξ, t) ∈ S×(0, T ),

and (11) implies that, for any t ≥ 0, Z t X ∂w 1 X (i, s) (w(i, s) − X(i, s)) (X(t, i) − w(t, i))2 = 2 ∂s 0 2 2 i∈Z Z

i∈Z Z

+ Ls (F (X(., s), s) + M(t).

Then, by using the first step, the result follows.

Proof of Theorem 3.1: Using the fact that u ˆ is a solution of (8) and ˆ for any t ≥ 0, we have X(t) ∈ K, X X ∂u ˆ fˆ(j, s)(X(j, s)−ˆ u(j, s)) ≤ 0 (i, s)(ˆ u(i, s)−X(i, s))+ ∂s 2 2

i∈Z Z

for any t ≥ 0.

j∈Z Z

Then, using the second part of Lemma 3.2 with u ˆ, we deduce (9).

At last, let us come back to the continuous model of Prigozhin (1) and assume that f ∈ BVloc (0, ∞; L2 (IR2 )) and there exists R > 0 such that spt(f ) ⊆ B(0, R),

(12)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

275

here B(0, R) denotes the a ball with center 0 and radius R. Thanks to,4 we know that (1) has a unique solution u in the sense that u ∈ 1,∞ Wloc (0, ∞; L2 (IR2 )), u(0) = 0 and, for any t ≥ 0, u(., t) ∈ K and Z f (t, x) − ∂t u(t, x) u(t, x) − ξ(x) ≥ 0 for any ξ ∈ K. IR2

For with the stochastic model, we rescale the source term tthexconnection . We set f , N N t x fˆ(t, i) = f , for any (t, i) ∈ [0, ∞) × ZZ 2 , N N

and we consider (X(t))t≥0 the Markov processus generated by fˆ as explained in the previous section. Theorem 3.2. Z IE

Under the assumption (12), we have 2 1 X(N t, [N x]) → 0 as N → 0. u(t, x) − N IR2

(13)

Proof: Let u ˆ be the solution of (8) and consider [N x] 1 uN (t, x) = u ˆ N t, [N x] and fN (t, x) = f t, N N

for any (t, x) ∈ [0, T ) × IR2 . It is not difficult to see that uN ∈ 1,∞ Wloc ((0, ∞); L2 (IR2 )) and, for any t ≥ 0, uN (., t) ∈ KN , where n 1o 1 for |x − y| ≤ KN = z ∈ L2 (IR2 ) ; |u(x) − u(y)| ≤ . (14) N N

In addition, for any ξ ∈ KN , we have R f (t, x) − ∂ u (t, x) u (t, x) − ξ(x) N t N N IR2 =

1 N3

where ˆ = N3 ξ(i)

P ˆ ˆ f(N t, i) − ∂ u ˆ (N t, i) u ˆ (N t, i) − ξ(i) , t i

Z

ξ(x) dx Ii

and

Ii = {z ∈ IR2 ; [N z] = i}.

ˆ and Now, since ξ ∈ KN and u ˆ is the solution of (8), then ξˆ ∈ K Z fN (t, x) − ∂t uN (t, x) uN (t, x) − ξ(x) ≥ 0 for any ξ ∈ KN . IR2

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

276

In other words uN is the solution of the nonlinear dynamic   ∂t uN + ∂IIKN (uN ) 3 fˆN 

(15)

u ˆN (0) = 0.

Thanks to,1 we know that in C([0, T ); L2(IR2 )).

uN → u

(16)

Indeed, ∩N ∈IN KN = K so that, letting N → ∞, we have ∂IIKN converges in the sense of graph to ∂IIK in L∞ (IR2 ). On the other hand, as N → ∞, in L2loc ((0, ∞); L2 (IR2 )).

fN → f

Then, by using classical perturbation result for nonlinear semigroup (cf.4 ), (16) follows. At last, thanks to Theorem 3.1, we have 2 R 1 IE IR2 N X N t, [N x] − uN (t, x) dx = IE = IE

h

h

≤

1 N4

≤

1 N2

≤

1 N

→0

1 N4 1 N4

P P

R Nt R 0

RtR 0

IR2

i |X N t, i − u ˆ N t, i |2 dx

i∈Z Z2

R Nt P 0

i |X N t, i − N uN (t, i)|2 dx

i∈Z Z2

i∈Z Z2

IR2

f

fˆ s, i ds

s [N x] N, N

dsdx

f τ, [NNx] dτ dx

as N → ∞.

so, by using (16), we deduce (13). Acknowledgements This work was partially supported by the French A.N.R. Grant JC05-41831 and by ”FLUPARTI” project (supported and funded by ”Conseil R´egional de Picardie”).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

277

References 1. F. Andreu, J. M. Mazon, J. D. Rossi and J. Toledo, The limit as p → ∞ in a nonlocal p−Laplacian evolution equation. A nonlocal approximation of a model for sandpiles. To appear in Cal. Var. Partial Diff. Equ. 2. G. Aronson, L. C. Evans and Y. Wu, Fast/Slow diffusion and growing sandpiles. J. Differential Equations, 131:304–335, 1996. 3. J. W. Barrett and L. Prigozhin, Dual formulation in Critical State Problems. Interfaces and Free Boundaries, 8, 349-370, 2006. 4. H. Br´ezis, Op´erateurs maximaux monotones et semigroups de contractions dans les espaces de Hilbert. (French). North-Holland Mathematics Studies, No. 5. Notas de Matem` atica (50). North-Holland Publishing Co., Amsterdam-London; American Elsevier Publishing Co., Inc., New York, 1973. 5. S. Dumont and N. Igbida, Back on a Dual Formulation for the Growing Sandpile Problem. To appear in Eur. Journal Appl. Math.. 6. L. C. Evans and F. Rezakhanlou, A stochastic model for sandpiles and its continum limit. Comm. Math. Phys., 197 (1998), no. 2, 325-345. 7. N. Igbida, Evolution Monge-Kantorovich equation. submitted. 8. L. Prigozhin, Variational model of sandpile growth. Euro. J. Appl. Math. , 7, 225-236, 1996

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

278

On radial solutions for Navier boundary eigenvalue problem with p-biharmonic operator Siham El Habib∗ and Najib Tsouli† University Mohamed I, Faculty of Sciences Department of Mathematics and Computer Oujda, Morocco E-mails: ∗[email protected], †[email protected] This work deals with the existence of radial solutions for a nonlinear eigenvalue problem involving the p-biharmonic operator. Keywords: p-biharmonic; Radial solutions.

1. Introduction Let us consider the nonlinear eigenvalue problem 2 ∆p u = λ|u|p−2 u inΩ, (Pp ) u = ∆u = 0 on∂Ω, where 1 < p < +∞, ∆2p u := ∆(|∆u|p−2 ∆u) is the operator of fourth order called p-biharmonic operator and Ω = B1 is the unit ball of IRN , N ≥ 2. The spectrum of the p-biharmonic operator was recently studied by many authors (cf.,2 ,3 ,4 ,5 ,6 ,7 ,8 and the references therein). 4 ˆ P. Dr´ abeck and M.Otani studied the spectrum of the problem (Pp ) in a bounded and smooth domain Ω of IRN (N ≥ 1). In one dimensional case and Ω = [0, 1],J. Benedikt (cf.2 ) gave the spectrum of the p-biharmonic operator under Dirichlet and Neumann boundary conditions. In his thesis, M. Talbi (cf.6 ) studied a variety of problems with pbiharmonic. He has included there many important techniques and tools concerning such problems. A. El Khalil, S. Kellati and A. Touzani (cf.5 ) showed that the spectrum of the p-biharmonic operator with weight and with Dirichlet boundary conditions contains at least one non-decreasing sequence of positive eigenvalues.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

279

Our study deals with the existence of radial solutions for the problem (Pp ). We would like to mention some works which were of main importance to achieve this work: The first one is of F. De Thelin9 where he showed that the problem −∆p u = λ|x|α |u|p−2 u, u ∈ W01,p (B1 ), u ≥ 0, u 6= 0,

(1)

has at least one positive radial solution. The second one is of A.Anane1 where he determined all the eigenvalues associated with a radial eigenfunction for the problem (1) and some properties of the eigenfunctions. We show similar results as those of A.Anane.1 We prove that the problem (Pp ) has at least a sequence of eigenvalues associated with a radial eigenfunction, any eigenvalue λn (n ≥ 1) associated to a radial eigenfunction is given according to the first eigenvalue λ1 and any radial eigenfunction associated to λn (n ≥ 1) has a finite number of components of the nodal set defined by {x ∈ Ω : u(x) 6= 0}. 2. Preliminaries In this section, we give some results and notations that will be used throughout this paper: For 0 < R < 1, we denote by: • BR = {x ∈ IRN : |x| < R}, • CR = {x ∈ IRN : R < |x| < 1}, • λ1,R (resp. µ1,R ) the first eigenvalue of the problem (P ) in BR (resp. CR ), • ϕR (resp ψR ) the eigenfunction associated with λ1,R (resp µ1,R ), • λ1 = λ1,1 . Definition 2.1. For any radial function u, we associate a real valued function u defined by u(x) = ϕ(|x|) where |x| is the Euclidian norm of x in IRN . The problem (Pp ) is equivalent to the following problem (for more details see4 ) (Pp0 ) where

∗ Find (v, λ) ∈ (Lp (Ω) \ {0}) × IR+ such that: f10 (v) = λg10 (v),

f1 (v) =

1 1 kvkpp , g1 (v) = kΛvkpp p p

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

280

and f10 (v) = Np (v), g10 (v) = Λ(Np (Λv)) Np design the Nemytskii operator defined by Np (v)(x) =

|v(x)|p−2 v(x) if 0 if v(x) = 0.

v(x) 6= 0

and Λ the inverse operator of −∆ : W 2,p (Ω) ∩ W01,p (Ω) → Lp (Ω). In the following lemma we give some properties of the operator Λ (cf .4 ): Lemma 2.1. (i) (Continuity): There exists a constant cp > 0 such that kΛf k2,p ≤ cp kf kp

holds for all p ∈ ]1, +∞[ and f ∈ Lp (Ω). (ii) (Continuity) Given k ∈ IN ∗ , there exists a constant cp,k > 0 such that kΛf kW k+2,p ≤ cp,k kf kW k,p

holds for all p ∈ ]1, +∞[ and f ∈ W k,p (Ω). iii) (Symmetry) The equality Z Z u · Λvdx Λu · vdx = Ω

Ω

p

p0

holds for all u ∈ L (Ω) and v ∈ L (Ω) with p ∈ ]1, +∞[. ¯ for all α ∈ ]0, 1[; (iv) (Regularity) Given f ∈ L∞ (Ω), we have Λf ∈ C 1,α (Ω) moreover, there exists cα > 0 such that kΛf kC 1,α ≤ cα kf k∞ .

¯ and f ≥ 0 (v) (Regularity and Hopf-type maximum principle) Let f ∈ C(Ω) 1,α ¯ then w = Λf ∈ C (Ω), for all α ∈ ]0, 1[ and w satisfies: w > 0 in Ω, ∂w ∂n < 0 on ∂Ω. (vi) (Order preserving property) Given f, g ∈ Lp (Ω) if f ≤ g in Ω, then Λf < Λg in Ω. We recall now the following result Proposition 2.1. Let X be a reflexive Banach space, M a weakly closed subset of X. Suppose φ : M −→ X is weakly lower semi-continuous on M then if φ is coercive on M , there exists u0 ∈ M such that φ(u0 ) = inf φ(v) v∈M

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

281

see.1 We close this section by the following results (for more details see4 ): Proposition 2.2. If u is a nontrivial solution of problem (Pp ) associated to an eigenvalue λ such that u does not change sign in Ω then λ = λ1 Proposition 2.3. If u is a solution of the problem (Pp0 ) then u ∈ C(Ω) Proposition 2.4. If f ∈ C(Ω) and f ≥ 0 then ω = Λf ∈ C 1,α (Ω) for all α ∈ (0, 1). ω satisfies ω > 0 in Ω and ∂ω ∂n < 0 on ∂Ω. 3. Existence of radial solutions The main objective of this section is to show that the problem (Pp ) in BR has an eigenvalue associated with a radial eigenfunction. Proposition 3.1. The problem (Pp ) when Ω = BR admits an eigenvalue λR associated with a radial positive eigenfunction. Proof. Let X be the completion of the space of radial functions of ID(Ω) under the norm of W 2,p (Ω) ∩ W01,p (Ω). Set Z |u|p dx = 1}, λR = inf{k∆ukpp ; u ∈ X and Ω

1 f (u) = k∆ukpp , p and M = {u ∈ X /

1 g(u) = p Z

Ω

Z

Ω

|u|p dx.

|u|p dx = 1}.

We verify first that f, X and M satisfies the conditions of the Proposition 2.1 then there exists u ∈ X , u 6= 0 such that ∀v ∈ X: Z Z |u|p−2 uvdx. (2) ∆(|∆u|p−2 ∆u)vdx = λR Ω

Ω

By using Green formula, we obtain Z Z p−2 |u|p−2 uvdx ∇(|∆u| ∆u)∇vdx = λR − Ω

(3)

Ω

In spherical coordinates ( r = |x|, r ∈ [0, R] and θ ∈ S = [0, 2π] × ((− π2 , π2 ))N −2 ) (3) becomes Z R Z R − rN −1 (|∆ϕ|p−2 ∆ϕ)0 φ0 dr = λR |ϕ|p−2 ϕφdr. (4) 0

0

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

282

Where u(x) = ϕ(r) and v(x) = φ(r), ∆u = ϕ00 (r) + Nr−1 ϕ0 (r) and ∆ϕ(r) is a notation to design the quantity ϕ00 (r) + N r−1 ϕ0 (r). We multiply (4 ) by ξ1 ∈ C ∞ (S) and after, we integrate on S. We obtain Z Z − ∇(|∆u|p−2 ∆u)∇v ξdx = λR |u|p−2 uv ξdx, (5) Ω

Ω

where ξ(x) = ξ1 (θ). Since u is radial then ∇(|∆u|p−2 ∆u)∇v ξ = ∇(|∆u|p−2 ∆u)∇(vξ). It follows that, for all v ∈ ID(Ω) Z Z |u|p−2 uvdx. |∆u|p−2 ∆u∆vdx = λR

(6)

Ω

Ω

Then by density, we conclude that λR is an eigenvalue of the problem (Pp ) in BR associated with u which is a radial eigenfunction. Remark 3.1. 1/ The problem (Pp ) is equivalent to the problem (Pp0 ) (see preliminaries section). 2/ If u is an eigenfunction of the problem (Pp ) associated with λR then v = −∆u is an eigenfunction associated with of (Pp0 ). R λR eigenvalue p p p 3/ λR = inf{kvkp ; v ∈ L (Ω); v radial and Ω |Λv| = 1} Proposition 3.2. The functionals ϕR and ψR are radial where ϕR (resp ψR ) is an eigenfunction associated to the first eigenvalue λ1,R in BR (resp. µ1,R in the crown CR ).

Proof. Let us show that ϕR is radial (a similarly proof is given for ψR ). For this end, we prove that u does not change sign in Ω. Since v = −∆u is an eigenfunction of the problem (Pp0 ) associated with λR then f1 (v) = λR g1 (v). From the p-homogeneity of f1 and g1 we deduce that f1 (v) − λR g1 (v) =

inf

u∈Lp (Ω)\{0}, u radial

f1 (v) − λR g1 (v).

(7)

Then f1 (v) − λR g1 (v) = 0 < f1 (|v|) − λR g1 (|v|). On the other hand, since |Λv| ≤ Λ|v| then f1 (v) − λR g1 (v) = 0 ≥ f1 (|v|) − λR g1 (|v|) ⇒ f10 (|v|) = λR g10 (|v|) i.e Np (|v|) = Λ(Np (Λ|v|)). It follows then that |v| > 0, i.e., v does not change sign on Ω and consequently u does not change sign on Ω. Proposition 2.2 allows us to conclude that λR = λ1,R . However, λ1,R is

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

283

simple then there exists t ∈ IR such that ϕR = tu and Consequently ϕR is radial. Now, we are going to study the regularity of radial solutions. We associate to ϕR and ψR the real valued functions ϕ˜R and ψ˜R such that: ϕR (x) = ϕ˜R (r) and ψR (x) = ψ˜R (r) where r = |x|. Proposition 3.3. (i) ϕ˜R ∈ C 1,α ([0, R]) for some α ∈ (0, 1). Moreover, ϕ˜0R (0) = 0, ∆ϕ˜R < 0 in ]0, R[ and ϕ˜R > 0 in [0, R[. (ii) ψ˜R ∈ C 1,α ([R, 1]) for some α ∈ (0, 1) and there exists r0 ∈ ]R, 1[ such 0 (r ) = 0. Moreover, ∆ψ˜ (r) < 0 in ]R, 1[. that ψ˜R 0 R (iii) The functionals ϕ˜R and ψ˜R are at least of class C 4 in ]0, R[. Proof. (i) Since ϕR does not change sign in BR then vR = −∆ϕR is an eigenfunction of (Pp0 ) associated with λ1,R which does not change sign in BR . Suppose that vR ≥ 0 then by Propositions 2.3 and 2.4 (cf.4 ), we deduce that vR ∈ C(Ω) then ϕ˜R ∈ C 1,α ([0, R]) for some α ∈ (0, 1), ϕ˜R > 0 on [0, R[ and ∆ϕ˜R < 0 on ]0, R]. Since ϕ˜R is radial then ϕ˜0R (0) = 0. (ii) Similarly to (i), we show that ψ˜R ∈ C 1,α ([R, 1]) for some α ∈ (0, 1). 0 (r ) = 0. Since ψ˜R (R) = ψ˜R (1) = 0 then there exists r0 ∈ ]R, 1[ such that ψ˜R 0 ˜ Moreover, ∆ψR < 0 in [R, 1]. (iii) Let us show that ϕ˜R is at least of class C 4 in ]0, R[. By using the spherical coordinates, we have (rN −1 (|∆ϕ˜R |p−2 ∆ϕ˜R )0 )0 = λ1,R rN −1 |ϕ˜R |p−2 ϕ˜R

(8)

where ∆ϕ˜R (r) is a notation to design ϕ˜00R (r) + Nr−1 ϕ˜0R (r). Since ϕ˜R is continuous then (|∆ϕ˜R |p−2 ∆ϕ˜R )0 is of class C 1 on ]0, R[, i.e ∆ϕ˜R is of class C 2 on ]0, R[, then we deduce that ϕ˜R ∈ C 4 (]0, R[). Similarly, we prove that ψ˜R ∈ C 4 (]0, R[). 4. Zeros points of eigenfunctions First, We prove the following lemma where we state the relation between eigenvalues and eigenfunctions of the problem (Pp ) in BR and those in the unit ball B1 . Lemma 4.1. (i) For all R ∈ ]0, 1], we have λ1,R = R−2p λ1 and ϕR (x) = 2p−N x R p ϕ1 ( R ) for all x ∈ BR . (ii) µ1,R is a real valued continuous function for all R ∈ ]0, 1[, strictly increasing and lim µ1,R = +∞.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

284

Proof. (i) Since ϕR is an eigenfunction of the problem (Pp ) associated with λ1,R when Ω = BR then Z Z p−2 |ϕR |p−2 ϕR vdx ∀v ∈ W 2,p (Ω)∩W01,p (Ω). |∆ϕR | ∆ϕR ∆vdx = λ1,R Ω

Ω

(9) x Let us Consider the change of variable x0 = R and set ϕR (x) = ϕ(x0 ) and v(x) = w(x0 ) then Z Z |∆ϕ|p−2 ∆ϕ∆wdx0 = λ1,R R2p |ϕ|p−2 ϕwdx0 , (10) B1

B1

2p

then λ1,R R is an eigenvalue of problem (Pp ) in B1 associated with ϕ. Since ϕ is positive then λ1 = λ1,R R2p and λ1,R = R−2p λ1 . 2p−N x ). Indeed, we have Let us now show that ϕR (x) = R p ϕ1 ( R Z Z |∆ϕR |p dx = |∆ϕ1 |p dx. (11) BR

B1

Since λ1 is simple then there exists t > 0 such that ϕ = tϕ1 . Hence Z Z |∆ϕ1 |p dx0 . |∆ϕR |p dx = RN −2p tp B1

BR

2p−N

x ). It follows then that tp = R2p−N Thus ϕR (x) = R p ϕ1 ( R (ii) For the proof of this assertion, one can consult the thesis of A. Anane .1

Now, we study the zero points of radial eigenfunctions. Lemma 4.2. If ϕ is a radial eigenfunction of the problem (Pp ) then N (ϕ) ˜ = {r ∈ [0, 1] such that ϕ(r) ˜ = 0} is finite and 0 ∈ / N (ϕ) ˜ where ϕ(x) = ϕ(|x|). ˜ Proof. ϕ is an eigenfunction of (Pp ) in B1 then ϕ(x) = 0 for all x: |x| = 1. Since ϕ is radial then 1 ∈ N (ϕ), ˜ therefore N (ϕ) ˜ 6= ∅. Set r = min{N (ϕ) ˜ \ {0}} since ϕ˜ does not change sign in ]0, r[ then 0 ∈ / N (ϕ). ˜ Let us show that N (ϕ) ˜ is finite. Indeed, suppose by contrary that there exists a sequence (rn )n ⊂ [0, 1] such that ϕ(r ˜ n ) = 0. For a subsequence still denoted by (rn )n , we can assume that limn rn = r for some r ∈ [0, 1] , rn 6= r, ∀n ∈ IN then: ϕ(r) ˜ = 0,

ϕ˜0 (r) = 0 and

ϕ˜00 (r) = 0.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

285

Hence N −1 0 ϕ (r) = 0 r which is in contradiction with (i) of Proposition 3.3. ∆ϕ(r) ˜ = ϕ00 (r) +

For n ≥ 1, we define a set Z(n) as follows :

Z(1) = {R11 } = {1}, Z(n) = {Rn1 , Rn2 , . . . , Rnn },

where (??)

Rnn = 1 i Rni = an−1 Rn−1 if 1 ≤ i ≤ n − 1 and n ≥ 2.

(an )n is a well defined sequence given by : 2p a1 µ1,a1 = λ1 a2p n µ1,an = µ1,an−1 if n ≥ 2. We verify easily that: 0 < a1 < a2 < . . . < an < . . . < 1, n−1 −2p λ1 , a2p n µ1,an = (Πi=1 ai )

0 < Rn1 < . . . < Rnn−1 = an < Rnn = 1, and

(

Rni = ai Rni+1 if 1 ≤ i ≤ n − 1 1 µ1,a n−1 Rn1 = Πi=1 ai = ( λn−1 )− 2p 1

Lemma 4.3. For any radial eigenfunction ϕ of the problem (Pp ) in B1 associated with an eigenvalue λ, we have N (ϕ) ˜ = Z(n) and λ = (Rn1 )−2p λ1 where n = card(N (ϕ)) ˜ and ϕ(x) = ϕ(|x|). ˜ Proof. We proceed by recurrence. For n = 1 we have N (ϕ) ˜ = {1} and ϕ˜ does not change sign in ]0, 1[ then λ = λ1 = (R11 )2p λ1 . Hypothesis: Suppose that for any radial eigenfunction u of (Pp ) in B1 such that card(N (˜ u)) = n we have N (˜ u) = Z(n) and λ = (Rn1 )−2p λ1 . ˜ = n+1 If φ is a radial eigenfunction of (Pp ) in B1 which satisfy card(N (φ)) ˜ = {T1 , T2 , . . . , T n, Tn+1 = 1}, we have and N (φ) R R |∆φ|p−2 ∆φ∆vdx = λ B1 |φ|p−2 φvdx B1 (12) ∀v ∈ W 2,p (B1 ) ∩ W01,p (B1 ).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

286

Then by considering the variable change x0 = xTn , one obtains Z Z p−2 0 2p |ω|p−2 ωvdx |∆ω| ∆ω∆vdx = λTn B Tn

B Tn

∀v ∈ W 2,p (BTn ) ∩ W01,p (BTn ),

(13)

where ω is the restriction of φ to BTn and ω(x0 ) = φ(x) then λTn2p is an eigenvalue of (Pp ) associated to ω in BTn . From the precedent hypothesis, we get λTn2p = (Rn1 )−2p λ1 . Similarly, the restriction of φ to CTn is an eigenfunction of constant sign on CTn associated with λ then µ1,Tn = λ and µ1,Tn Tn2p = (Rn1 )−2p λ1 . We conclude that Tn = an , Tn−1 = an−1 = Rnn−1 = an−1 Rnn . Consequently, ˜ = {R1 , . . . , Rn+1 } = Z(n + 1). N (φ) n+1 n+1 5. Main result Set ϕ˜1 (|x|) = ϕ1 (x) and ψ˜ai (|x|) = ψai (x) for all x ∈ B1 . It’s easy to see that ϕ˜01 (1) 6= 0 and ψ˜a0 i (1) 6= 0. Set ( ϕ1 ( Rx1 ) if |x| ≤ Rn1 n un (x) = x ) if Rni ≤ |x| ≤ Rni+1 , i < n and n ≥ 2 δi ψai ( Ri+1 n

where δi is defined by    δ1 =

ϕ˜01 (1) ˜0 (a1 ) a1 ψ a1

˜0

ψa (1)   δi = δ1 × Πij=2 ( ˜j−1 ) if i 6= 1. a ψ 0 (a ) j

aj

j

Let us consider the sequence (λn )n defined by λn = (Rn1 )−2p λ1 . Theorem 5.1. (i) For all n ≥ 1, un is a radial eigenfunction of problem (Pp ) in B1 associated to the eigenvalue λn . (ii) If u is a radial eigenfunction of problem (Pp ) in B1 associated with an eigenvalue λ then there exists n ≥ 1 and t ∈ IR such that: λ = λn and u = tun . Proof. (i) If n = 1 then λn = λ1 and un = ϕ1 . The result holds. If n ≥ 2. Let v be the restriction of un to BR1n and wi the restriction of un Ri

n to the crown CRi+1 = {x ∈ IRN : Rni < |x| < Rni+1 }. n Firstly, since v(x) = ϕ1 ( Rx1 ) then by lemma 4.1, v is an eigenfunction of n (Pp ) in Ω = BR1n associated with λ1,R1n = (Rn1 )−2p λ1 = λn .

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

287

Secondly, since ψai is an eigenfunction of (Pp ) in Ω = Cai = {x ∈ IRN : ai < |x| < 1} associated to µ1,ai then Z Z |ψai |p−2 ψai vdx |∆ψai |p−2 ∆ψai ∆vdx = µ1,ai (14) C ai C ai 1,p 2,p ∀v ∈ W (Cai ) ∩ W0 (Cai ).

Set x0 = Rni+1 x and wi (x0 ) = ψai (x) then we obtain Z Z p−2 0 i+1 −2p ) µ |∆w | ∆w ∆vdx = (R i i 1,ai n Ri Ri C

n i+1 Rn

∀v ∈ W

C

2,p

Rin (CRi+1 ) n

∩

Rin W01,p (CRi+1 ). n

n i+1 Rn

|wi |p−2 wi vdx0

(15)

Ri

n This shows that wi is an eigenfunction of (Pp ) in CRi+1 associated with n

(Rni+1 )−2p µ1,ai . However, since Rni+1 = a1i Rni = (Πij=1 aj )−1 Rn1 then 1 . From where , we con(Rni+1 )−2p = (Rn1 )−2p (Πij=1 aj )2p = (Rn1 )−2p µλ1,ai Ri

n clude that wi is an eigenfunction on (Pp ) in CRi+1 associated with λn . n To complete the proof, we need to study the regularity of un in the neighborhood of the points x which verify |x| = Rni . We prove that un is of class C 4 at x: |x| = Rni . Set v˜(|x|) = v(x) and w˜i (|x|) = wi (x). Since v˜(Rn1 ) = w˜i (Rni ) = 0 ∀1 ≤ i ≤ n − 1 then un is continuous at x: |x| = Rni . The function un is of class C 1 at x : |x| = Rni . Indeed, w 0˜ (Ri ) i i ψ˜0 (ai ) and w0˜ (Ri ) = δi+1 ψ˜0 (1) then i−1 n = 1 i.e w˜0 (Ri ) = δi+1

i

n

Rn

ai

i−1

n

ai

Rn

0˜ (Ri ) wi−1 n

w˜i0 (Rin )

= w˜i0 (Rni ). We can also easily verify that v˜10 (Rn1 ) = w˜10 (Rn1 ) . The result then is proved. We have N − 1 ˜0 i wi (Rn ) (16) ∆w˜i (Rni ) = w˜i00 (Rni ) + Rni =0 (17) = ∆wi−1 ˜ (Rni+1 ) (18) N −1 N −1 i i i i 00 ˜ (R ) = − i w0˜ (R ) what imthen w˜i00 (Rn ) = − Ri w˜i0 (Rn ) and wi−1 n n i−1 Rn n 00 ˜ (Ri ). plies that w˜i00 (Rni ) = wi−1 n Similarly, we prove that v˜00 (R1 ) = w00˜ (R1 ). It follows then that un is of i

n

class C 2 at x : |x| = Rni . un is of class C 3 at x : |x| = Rni : w˜i and v˜ verifies the equations:

i−1

rN −1 (|∆w˜i (r)|p−2 ∆w˜i (r))0 = λn

n

Z

r 0

sN −1 |w˜i |p−2 w˜i ds

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

288

rN −1 (|∆˜ v (r)|p−2 ∆˜ v (r))0 = λn

Z

r 0

then r

N −1

0

(p − 1)(∆w˜i ) = λn |∆w˜i (r)|

2−p

rN −1 (p − 1)(∆˜ v )0 = λn |∆˜ v (r)|2−p

Z Z

sN −1 |˜ v |p−2 v˜ds r 0 r 0

sN −1 |w˜i |p−2 w˜i ds sN −1 |˜ v |p−2 v˜ds

1st case: if 1 < p < 2 then at x : |x| = Rn1 , we have (∆˜ v )0 |r=R1n = 0 this implies that v˜(3) (Rn1 ) +

N − 1 ˜00 1 N − 1 ˜0 1 v (Rn ) − v (Rn ) = 0 Rn1 Rn1

˜0 1 However, v˜00 (Rn1 ) = − NR−1 ˜(3) (Rn1 ) = − RN1 v˜00 (Rn1 ) . 1 v (Rn ) then v n n Since v˜00 is continuous at R1 then v˜(3) is continuous at R1 . The same proof n

n

is given for w˜i . ˆ 2nd case: if p > 2. By Dr` abeck and Otani (cf.4 ), if u(p) is a solution of the problem (Pp ) associated to the eigenvalue λ(p) then λ(p0 ) defined 1 1 by (λ(p0 )) p0 = (λ(p)) p is an eigenvalue of the problem (Pp0 ) associated 1 with the eigenfunction u(p0 ) defined by u(p0 ) = λ(p) |∆u(p)|p−2 ∆u(p) where p 0 p = p−1 . We have 1 < p0 < 2 then similarly to the first case, we obtain v˜(3) (p0 ) is 1 continuous at Rn1 . Since v˜(p0 ) = λ(p) |∆˜ v |p−2 ∆˜ v , we deduce that v˜(3) (p) is 1 continuous at Rn . Consequently, un is of class C 3 at x: |x| = Rni . un is of class C 4 at x: |x| = Rni is proved similarly as for the assertion: ” un is of class C 3 ”. The proof then is completed. (ii) To prove this assertion, let u be a radial eigenfunction of (Pp ) in B1 associated to an eigenvalue λ then by the preceding results we have λ = λ1 (Rn1 )−2p = λn where n = card(N (˜ u)) . The restriction v (resp. wi ) Ri

Ri

n n of u to BR1n (resp. CRi+1 ) is an eigenfunction of (Pp ) in BR1n (resp. CRi+1 ) n

Ri

n

n which does not change its sign in BR1n (resp. CRi+1 ). n Since λ is simple and from (i) , we deduce that: λ = λn and there exists t ∈ IR such that u = tun .

References 1. A. Anane, Etude des valeurs propres et de la r´esonance pour l’op´erateur pLaplacien, th`ese de Doctorat, (Universit´e Libre de Bruxelle, 1988).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

289

2. J. Benedikt, On the Discreteness of the Dirichlet and Neumann p-Biharmonic problem, (Abst. Appl. Anal., Vol. 293(2004)), No 9, pp. 777-792. 3. J. Benedikt, Uniqueness theorem for the p-biharmonic Equations, (Electron. J. Differential Equations 2002), No 53, pp. 1-17. ˆ 4. P. Dr` abek and M. Otani, Global Bifurcation Result for the p-Biharmonic Operator, (Electronic Journal Of Differential Equations 2001), No 48, pp. 1-19. 5. A. El Khalil, S. Kellati and A. Touzani, On the spectrum of the p-Biharmonic Operator, 2002-Fez Conference On partial differential Equations, (Electronic Journal of Differential Equations, Conference 09, 2002), pp. 161-170 6. M. Talbi, Etude des probl`emes elliptiques d’ordre quatre gouvern´ees par l’ op´erateur p-biharmonique, Th`ese de doctorat, (Facult´e des sciences, Oujda, Maroc, 2007). 7. M. Talbi and N. Tsouli, Existence and uniqueness of a positive solution for a non homogeneous problem of fourth order with weight, 2005- Oujda International Conference on Nonlinear Analysis, (Electronic Journal of Differential Equations, Conference 14, 2006), pp. 231-240. 8. M. Talbi and N. Tsouli, On the Spectrum of the weighted p-Biharmonic Operator with weight, (Mediterr. J. Math. 4 (2007)). 9. F. De Th´elin, Quelques r´esultats d’existence et non existence pour une E.D.P elliptique non lineaire, (C. R. Acad. Sc. Paris 1984), N 18 pp. 911-914.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

290

Floquet states of periodically time-dependent harmonic oscillators Y. Achkar, S. Sayouri, and A. L. Marrakchi Laboratoire de Physique Th´ eorique et Appliqu´ ee (LPTA) D´ epartement de Physique Facult´ e des Sciences Dhar-Mehraz B.P. 1796, F´ es-Atlas, F´ es, Morocco Floquet theory together with the resonating averages method (RAM) were used to solve the Schr¨ odinger equation of periodically-time-dependent harmonic oscillators. The approach gives another way of solving this equation, and lead to identical analytical solutions to those published in the literature based on different other methods. Keywords: Time-dependent harmonic oscillator; Floquet operators; Floquet states; Jump operators; Resonating averages method; Uncertainty relation.

1. Introduction Numerous physical and mathematical studies have been devoted to the construction of consistent formalism of explicitly time-dependent quantum systems1 -.9 Among these quantum systems, the time-dependent harmonic oscillator has attracted considerable interest in the past few decades3 -,1218 ,25 not only because it is a system that can be exactly solved and is considered as a great pedagogical tool, but also because it is a very relevant system and possessing many applications in different areas of physics: quantum optics, plasma physics, quantum field theory, gravitation, cosmology, etc. The harmonic oscillator with a time-dependent mass and a constant frequency has been the subject of many researches16-,1921 -.25 Abdalla et al.18 have studied this system in order to describe the electromagnetic field intensities in a Fabry-Perot cavity. The harmonic oscillator with a constant mass and a time-dependent frequency was used by Lemos et al.20 in a theoretical approach of the expansion of the universe and by Paul23 for the

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

291

discovery of electromagnetic traps for charged and neutral particles (known as Paul-traps). In previous work,24 we have introduced a direct theoretical approach for solving the equations of motion of the time-dependent harmonic oscillator, in Heisenberg picture, with some direct applications. The purpose of this paper is to solve the Schr¨ odinger evolution equation of periodically-time-dependent harmonic oscillator. We have established a theoretical approach based on the Floquet theory 1 combined with the resonating averages method elaborated by Lochak and Thiounn.2 Hence,by using the Floquet decomposition operators and developing the evolution operator with the help of the resonating averages method from first order to second ameliorated order, we have determinate the Floquet operators, and Floquet states. Moreover, we have proposed a theoretical form of jump operators between the instantaneous Floquet states, and verified the uncertainty principle. This paper is organized as follows: in the first part, we present the theoretical formalism of the purposed approach. In the second part, we apply the method for the forced harmonic oscillator, the harmonic oscillator with a time-dependent mass and a constant frequency, the harmonic oscillator with a constant mass and a time-dependent frequency, and the harmonic oscillator with a time-dependent mass and frequency. Some direct comparisons of our results with some published works are given,3 ,59 -,1218 2. Formalism We consider a quantum system described by a time-dependent Hamiltonian such as H(t) = H0 + µH1 (t)

(1)

where H0 is the Hamiltonian of the unperturbed oscillator, and H1 (t) the Hamiltonian of perturbation with amplitude µ. 2.1. Floquet approach In the case of periodically time-varying Hamiltonian, the Floquet theorem asserts the existence of an operator V (t), a fundamental solution of the well-known Schr¨ odinger evolution equation, in the form1 V (t) = T (t)e−iRt/~

(2)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

292

Where T (t) is a periodic unitary operator of the same period as H1 (t), and R is a constant hermitian operator. The time-evolution operator can be written as U (t, t0 ) = V (t)V −1 (t0 )

(3)

in the interaction picture it satisfies the following differential equation i~

dUI (t, t0 ) = µHI (t)UI (t, t0 ) dt

(4)

where HI (t) = eiH0 t/~

(5)

According to the Floquet theorem, the Floquet states, solutions of the Schr¨ odinger equation, are defined as |ψn (t)i = T (t)|φn (t)i

(6)

where |φn (t)i are the eigenstates of R, with the eigenvalues εn , such as i~

d|Φn (t)i = R|Φn (t)i dt

(7)

and |φn (t)i = e−iεn t/~ |ni

(8)

|ni are the number states of the unperturbed system. We notice that the fundamental propriety of those Floquet states is the fact that they form a complet set of time-dependent solutions in the Hilbert space. So, the global state |Ψ(t)i of the quantum system can be written in the form of a linear combination of the |Ψn (t)i : X |Ψ(t)i = an |Ψn (t)i n

where the an are time-independent coefficients. 2.2. Jump operators

We introduce the operators A(t) and A+ (t) which are responsible for the instantaneous transitions between the Floquet states. Their effects on the Floquet states are similar to those of the conventional annihilation and creation operators. These operators must satisfy the conditions3 √ A(t)|ψn (t)i = n|ψn−1 (t)i (9)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

293

A+ (t)|ψn (t)i =

√

n + 1|ψn+1 (t)i

(10)

So, their expressions are defined by the relations A(t) = ei A+ (t) = e

∆ε ~ t

T (t)aT + (t)

−i ∆ε ~ t

T (t)a+ T + (t)

(11) (12)

where ∆ε is the difference between two successive Floquet levels: ∆ε = εn+1 − εn

(13)

The Floquet decomposition (Eq. (2)) gives us different possibilities to solve the evolution equation of the periodically-time-dependent harmonic oscillator. However, the determination of the corresponding Floquet operators is not unique, while the Floquet theory supposes the existence of an infinite number of couples (R, T (t)), solutions of equation (2), depending on the form of the perturbed Hamiltonian. To determine these two operators, we used the technique of the resonating averages method elaborated by G.Lochak,2 based on the generalized Bogoliubov approximation. 2.3. Resonating averages method (RAM) The principle of this method consists in the separation of the perturbed Hamiltonian HI (t), written in the interaction picture (Eq. (5)), into an fI (t), such as : averaging part, HI (t), and an oscillating part H fI (t) dH dt

(14)

(HI )k eiwk t

(15)

eiwk t iwk

(16)

HI (t) = HI (t) +

with HI (t) =

n−1 X k=0

and fI (t) = H

∞ X

k=n

(HI )k

where (HI )k are a set of constant hermitian operators, and wk a frequency sequence from which one can extract a subsequence which contains the frequency w0 = 0. The RAM stipulates that if one supposes that this subsequence contains a set of a finite number n of frequencies as {w0 , w1 , ..., wn−1 }, the other frequencies {wn , wn+1 , ...etc} are not harmonic or harmonic combinations of the n first frequencies.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

294

2.3.1. First order and first order ameliorated solutions The application of the RAM to Eq. (4), to first order of µ, gives the differential equation i~

d(1) VI (t) = µHI (t)(1) VI (t) dt

With the initial condition (1)

(1)

(17)

VI (t0 ) = 1, one has

UI (t) =(1) VI (t) =(1) T (t)e−i

(1)

Rt/~

The first order ameliorated solution is defined as iµ f (1a) (1) UI (t) = [1 − H VI (t) I (t)] ~

(18)

(19)

2.3.2. Second order and second order ameliorated solutions The second order fundamental solution of Eq. (4) is defined by the operator (2)

UI (t) = [1 −

iµ f HI (t)]Γ(t) ~

(20)

where Γ(t) is an unitary operator solution of : i~

iµ2 ZI (t) dΓ(t) = {µHI (t) + [SI (t) + ]}Γ(t) dt ~ 2

(21)

where fI fI HI − HI H ST (t) = H

f f fI dHI − dHI H fI ZI (t) = H dt dt One may define the second order ameliorated solution as (2a)

UI (t) = [1 −

iµ f HI (t) + µ2 A2 (t)]Γ(t) ~

(22) (23)

(24)

where the operator A2 (t) is given by A2 (t) =

f2 (t) fI (t) H 1 f Z I { S (t) + − } I ~2 2 2

(25)

Determination of first and second order ameliorated solutions (1a) U1 (t) and (2a) U1 (t) (Eqs. (19) and (24)) enables to obtain the first and second order couples ((1a) R,(1a) T (t)), ((2a) R,(2a) T (t)) and Floquet-states |(1a) ψn (t)i, |(2a) ψn (t)i, respectively.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

295

3. Applications 3.1. Forced harmonic oscillator (FHO) We have applied the above method, and compared our results to those of other studies,39 -12 which have used other methods, to a FHO with the following Hamiltonian H(t, ν0 ) =

p2 1 + m0 ω02 q 2 + µ sin(ν0 t)q 2m0 2

(26)

where m0 and ω0 are the constants describing the mass and the frequency of the unperturbed oscillator, respectively. By introducing the annihilation operator a p 1 p a = √ ( β0 q + i √ ) (27) β0 2~

where

β0 = m 0 ω 0

(28)

the quantized form of H(t) is 1 H(t, ν0 ) = ~ω0 (a+ a + ) + µ 2

s

~ (a + a+ ) sin ν0 t = H0 + µH1 (t) (29) 2β0

We applied the RAM to the perturbed Hamiltonian HI (t), expressed in the interaction picture as Eq. (5), choosing a frequency sequence such as ωk = 0, ν0 − 2ω0 , −(ν0 − 2ω0 )

(30)

we deduced from Eq. (29) the averaging and oscillating parts of HI (t), given respectively by HI (t) = 0 and

Then

where

fI (t) dH = dt

s

~ sin ν0 t(e−iω0 t a + eiω0 t a+ ) 2β0

fI (t) = α0 (t)a + α∗0 (t)a+ H −1 α0 (t) = 2

s

e−i(ν0 +ω0 )t ~ ei(ν0 −ω0 )t ( + ) 2β0 ν0 − ω0 ν0 + ω 0

(31)

(32)

(33)

(34)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

296

3.1.1. First order calculations The fundamental first order ameliorated solution of Eq. (4) is obtained as (1a)

U (t) = {1 −

H0 t iu (α1 (t)a + α∗1 (t)a+ )}e−i ~ ~

(35)

where α1 (t) = eiω0 t α0 (t)

(36)

From Eq. (2), we deduced the Floquet operators such as (1a)

T (t) = 1 −

iu (α1 (t)a + α∗1 (t)a+ ) ~

(1a)

R = H0

(37) (38)

The ameliorated first order operator (1a) R is simply equal to the unperturbed Hamiltonian H0 . This absence of first order linear effects in is one of the characteristic properties of the Floquet Hamiltonian. This fact follows directly from the perturbative approach applied to the unitary operator (Eq. (19)). However, it was shown that the phenomenon is deeper for a large class of field pulses which are not necessarily monochromatic, and all odd perturbative corrections must vanish13 -,1426 Then, the eigenstates of the Floquet operator (1a) R are obtained as |(1a) Φn (t)i = e−i

(1a) Rt ~

1

|ni = e−i(n+ 2 )ω0 t |ni

(39)

and the Floquet states are given by the following expression (1a) ε t √ √ iµ n |(1a) Ψn (t)i = e−i ~ {|ni − [α1 (t) n|n − 1i + α∗1 (t) n + 1|n + 1i]} ~ (40) where 1 (1a) (41) εn = ~ω0 (n + ) 2 To determine the expectation values and the uncertainty relation of the position and impulsion operators, one can write s ~ (a + a+ ) (42) q= 2β0 p=

r

~β0 + (a − a) 2

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

297

The first ameliorated order expectation values of q, p, q 2 and p2 in the Floquet states, given by Eq. (40), are

h(1a) q 2 i =

h(1a) qi =

−µ sin ν0 t m0 (ν02 − ω02 )

(43)

h(1a) pi =

−µν0 cos ν0 t ν02 − ω02

(44)

~ µ2 {2n + 1 + 2 [2n(n + 1)Re(α21 ) + ((2n + 1)2 + 2)|α1 |2 ]} (45) 2β0 ~

µ2 ~β0 {2n+1+ 2 [−2n(n+1)Re(α21 )+((2n+1)2 +2)|α1 |2 ]} (46) 2 ~ The explicit expression of the uncertainty product is h(1a) p2 i =

(1a)

h∆q∆pi =

~ (2n + 1) (2n + 1){1 + µ2 (ν 2 cos2 ν0 t + ω02 sin2 ν0 t)}1/2 2 ~β0 (ν02 − ω02 )2 0 (47)

3.1.2. Second order calculations To solve the differential equation (21), we begin by expressing the operator Z(t) in the form dα∗0 (t) ) dt Then, we deduce its averaging and oscillating parts as Z(t) = 2Im(α0 (t)

Z(t) =

−i~ 2m0 (ν02 − ω02 )

(48)

(49)

e i~(e2iν0 t + e−2iν0 t ) dZ(t) = dt 4m0 (ν02 − ω02 )

(50)

i~dΓ(t) iµ2 = Z(t)Γ(t) dt 2~

(51)

Because HI = 0 and SI = 0,Eq. (21) is reduced to

So, we deduce Γ(t) = e

−iµ2 t 4~m0 (ν 2 −ω2 ) 0 0

Then, the second order evolution operator is given by H0 t iu (2) U (t) = {1 − (α1 (t)a + α∗1 (t)a+ )}e−i ~ Γ(t) ~

(52)

(53)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

298

and the corresponding Floquet components are (2)

1 µ2 R = ~ω0 (a+ a + ) + 2 4m0 (ν02 − ω02 )

(54)

iu (α1 (t)a + α∗1 (t)a+ ) =(1a) T (t) ~ The ameliorated second order solution is iu f (2a) 2 UI (t) = {1 − H I (t) + µ A2 (t)}Γ(t) ~ with f2 (t) e −H Z(t) 1 e I A2 (t) = 2 (S(t) + ) ~ 2 By integrating Eq. (50), we get (2)

T (t) = 1 −

2iν0 t − e2iν0 t ) e = ~(e Z(t) 8m0 ν0 (ν02 − ω02 )

then,

(55)

(56)

(57)

(58)

2

u ∗ + U (t) = {1 − iu ~ (α1 (t)a + α1 (t)a ) − 2~2 (ϕ(t) (2a) 2 2 ∗2 +2 2 + +α1 (t)a + α1 (t)a + |α1 | (2a a + 1))}e−i Rt/h

(2a)

(59)

where ϕ(t) =

−i~ sin ν0 t cos ν0 t 2m0 ν0 (ν02 − ω02 )

(60)

The ameliorated second order Floquet-Hamiltonian has the following form (2a)

1 µ2 R = ~ω0 (a+ a + ) + 2 4m0 (ν02 − ω02 )

∗ + T = 1 − iu ~ (α1 (t)a + α1 (t)a ) − ∗2 +2 2 + +α1 (t)a + |α1 | (2a a + 1))

(2a)

u2 2~2 (ϕ(t)

+ α21 (t)a2

(61)

(62)

In the same way, we determined the ameliorated second order Floquet states |(2a) Ψn (t)i =(2a) T |(2a) Φn (t)i

(63)

where |(2a) Φn (t)i = e−i

(2a)

Rt/~

|ni = e−i

(2a)

εn t/~

|ni

(64)

and the quasi-energy is (2a)

1 µ2 εn = ~ω0 (n + ) + 2 4m0 (ν02 − ω02 )

(65)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

299

The explicit expression of the Floquet states is 2

µ 2 |(2a) Ψn (t)i = {[1 + ϕ(t) − 2~ 2 (2n + 1)|α1 | ]|ni √ √ iµ ∗ − ~ [α1 (t) n|n − 1i + α1 (t) n + 1|n + 1i] 2

µ 2 − 2~ 2 [α1 (t)

p

n(n − 1)|n − 2i + α∗2 1 (t)

p

(n + 1)(n + 2)|n + 2i]}e−i

(66) (2a) ε t n ~

Let us notice that the expression of the ameliorated second order Floquet operator (2a) T (t), given by Eq. (62), is comparable to the so-called coherent states generator published by Breuer et al.,3 and by Fox et al.,9 as well as the ameliorated second order Floquet states. The detailed comparisons with these works accompanied with some discussions are given in section 3.1.4. Moreover, the expression of the quasi-energy (Eq. (65)) is identical to that given in some other works,39 -13 which supports the observations previously made. Consequently, the corrections are absent to first order but appear, when the approximations are pushed to second order, and are expressed by a shift Dε given by Dε =

µ2 4m0 (ν02 − ω02 )

(67)

which does not depend on the quantum number n, but on the strength µ and the pulsation ν0 of the interacting field. This means effect of the perturbation is as a whole spectral shift of all Floquet levels. This ∆ε is giving by ∆ε = εn+1 − εn

(68)

The sense of displacement depends on the sign of Dε .If Dε > 0 , the levels shift up and inversely they shift down if Dε < 0. Therefore, the frequency of a transition of the Floquet states, which corresponds to an absorption or emission of the so-called ”Floquet quanta”,14 is equal to that of the stationary states. So, we can write (2a)

∆ε = ~(2a) ωF

(69)

and (2a)

ωF = ω 0

(70)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

300

The corresponding expectation values of the position and momentum operators are given by s ~ iµ µ2 (2a) h qi = {2Im(α1 ) + 2 (2n + 1)[−ϕ(t)Re(α1 ) + |α1 |2 Im(α1 )]} 2β0 ~ ~ (71) s µ2 ~ µ {2Re(α1 ) + 2 (2n + 1)[−ϕ(t)Im(α1 ) + |α1 |2 Re(α1 )]} h(2a) pi = − 2β0 ~ ~ (72) h(2a) q 2 i =

~ 4µ2 µ4 |ϕ(t)|2 {(2n+1)− 2 Im(α1 )+ 4 (2n+1)[ +2ϕ(t)Re(α1 )Im(α1 ) 2β0 ~ ~ 4

(73) +

5(2n2 + 2n + 3) 4 (2n2 + 2n + 7) Re (α1 ) + Im4 (α1 ) 4 4 −

h(2a) p2 i =

(6n2 + 6n + 11) 2 Re (α1 )Im2 (α1 )]} 2

4µ2 µ4 |ϕ(t)|2 ~β0 {(2n + 1) + 2 Re(α1 ) + 4 (2n + 1)[ − 2ϕ(t)Re(α1 )Im(α1 ) 2 ~ ~ 4

(74) +

(2n2 + 2n + 7) 4 5(2n2 + 2n + 3) Im4 (α1 ) + Re (α1 ) 4 4

(6n2 + 6n + 11) 2 Re (α1 )Im2 (α1 )]} 2 and the explicit uncertainty product is −

h(2a) ∆q∆pi = +

~ µ4 sin2 2ν0 t (2n + 1){1 + 2 2 2 [ 2 8~ m0 (ν0 − ω02 )2 2ν02

(75)

3(2n2 + 2n + 1) 2 2 (ω0 sin ν0 t + ν02 cos2 ν0 t)]} ω02 (ν02 − ω02 )2

3.1.3. Jump operators In section 2.2, we have defined the operators A(t) and A+ (t) which are called jump operators,315 From Eqs. (11) and (12), we can write, to first order in µ A(t) = ei

(1a)∆ε ~

t (1a)

T (t)a(1a) T + (t)

(76)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

301

A+ (t) = e−i

(1a)∆ε ~

t (1a)

T (t)a+(1a) T + (t)

By substituting the expression (37) of the unitary operator obtain the two conjugal operators such as, A(α1 , t) = a +

iµ ∗ α (t) ~ 1

A+ (α1 , t) = a+ −

iµ α1 (t) ~

(77) (1a)

T (t) , we

(78)

(79)

The action of those ladder operators on the Floquet states |( 1a)ψn (t)i is equivalent to the action of the annihilation and creation operators on the stationary states. These operators are time dependent in the Schr¨ odinger picture. Their expressions (78) and (79) are exactly identical to those published by Breuer et al.3 The operators A(t) and A+ (t) satisfy the commutation relation [A(t), A+ (t)] = 1 One can also give an hermitical symmetric operator15 N (t) such as N (t) =

1 (AA+ + A+ A) 2

(80)

From the definitions A(t)|ψn (t)i = A+ (t)|ψn (t)i =

√

n|ψn−1 (t)i

(81)

√ n + 1|ψn+1 (t)i

(82)

we have A+ A|ψn (t)i = n|ψn (t)i

(83)

AA+ |ψn (t)i = (n + 1)|ψn (t)i

(84)

and

Then, the exact action of the operator N (t) on the Floquet states is defined as 1 N (t)|ψn (t)i = (n + )|ψn (t)i 2

(85)

This operator N (t) is equivalent to the number operator of the unperturbed harmonic oscillator.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

302

3.1.4. Comparisons and discussions We have presented in the above paragraphs some analytical results of Floquet operators, Floquet states, expectations values and uncertainties products of the position and momentum operators. To test the validity of our approach and to confirm the accuracy of our calculations, we will give some direct comparisons with other works given in the literature,39 -12 . In particular, Breuer et al.,3 have generalized the stochastic wave-function method to the description of open dissipative systems in strong Laser field, and have employed the Floquet representation for quantum systems with timeperiodic Hamiltonian, among which the forced harmonic oscillator, submitted to a Laser field with an Hamiltonian of the same form of Eq. (26). Fox et al.9 have presented a non perturbative method based on a quasiadiabatic method. This approach has been applied to an harmonic oscillator in an electric field of the form, r ~ EF (t) = −eE0 cos ωt(a + a+ ) (86) 2m0 ω0 To compare our results to those of these authors, we should make some transformations by using the trigonometric formulation cos x = sin(x + Π 2 ).Then, in all our trigonometric functions (sinus or cosines) we add a Π phase (+ Π2 ) i.e. ( our results −→phase+ 2 Fox results) i)Annihilation and position operators in the Heisenberg picture: Breuer et al.3 have calculated the expression of annihilation and position operators, for the FHO in the Heisenberg picture, to first order in µ. Therefore, we shall use the evolution operator obtained in Eq. (35) to express aH (t) and qH (t).So, we can write aH (t) =(1a) U + (t)a(1a) U (t)

(87)

qH (t) =(1a) U + (t)q (1a) U (t)

(88)

and

we obtain 0

aH (t) = ae−iω0 t +

eiν0 t e−iν0 t iµ ( + ) 2 ν0 + ω 0 ν0 − ω 0

(89)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

303

and qH (t) =

s

0 eiν0 t − e−iν0 t ~ {ae−iω0 t + a+ eiω0 t + iµ ω0 } 2β0 ω02 − ν02

(90)

with 0

µ =√

µ 2~β0

(91)

These expressions are identical to those, aH (t, t0 ) and xH (t, t0 ), published by Breuer et al.3 using the initial condition t0 = 0 in the integration of the f e operator dHdtI (t) in Eq. (32) and putting H(0) = 0 and ~ = 1 With those conditions, we can obtain exact expressions of aH (t, t0 ) and xH (t, t0 ) given in reference.3 But in our approach, we need to have general forms without imposing any initial condition on our quantum system. ii)Floquet operator and Floquet states Let us notice that our ameliorated first order operator (1a) T (t) and ameliorated second order one (2a) T (t) correspond to a first and a second order development of the Breuer’s operator eϕ(t) D(w) and the Fox’s operator eϕF (t) DF (η), respectively (but for this latter we must take the trigonometric transformation previously indicated). Where the so-called coherent state generator3 is defined by D(w) = exp(wa+ − w∗ a)

(92)

By comparing the Breuer’s variable w(t) with our variable α1 (t) of Eq. (36), and the Fox’s one η(t), we have the equality w(t) =

ν0 t+ π −iµ ∗ α1 (t)(Eq.(116)of [3]) −→2 the F ox0 s variable η(t) (Eq.(35)of [9]) ~

(93) as well as the phase ϕ(t) obtained in Eq. (60) is directly comparable to the Breuer’s one ϕB (t) (see the Eq. (120) of reference3 ) and exactly the same with the Fox’s one ϕF (t) (see Eq. (34) of reference9 ). iii)Quasi-energy The Floquet energy given by Eq. (65) is identical to that calculated by Breuer et al.,3 Fox et al.,9 Lefebvre et al.,10 Maitra et al.11 and Samal et al.12

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

304

iv)Expectation values and uncertainty products The expectation values of the position q and momentum p operators are exactly comparable to those given to first order in µ by Fox et al.,9 but to second order, our values contain a second order correction term in µ . Finally, our uncertainty product contains a 2nd order correction term in µ for the case of the ameliorated first order uncertainty relation, while in the case of ameliorated second order one, it contains a 4th order correction term in µ. v)Jump operators Let us notice that the expressions of jump operators given by Eqs. (78) and (79) are identical to those published by Breuer et al.3 3.2. Pulsating Harmonic Oscillator An other good candidate for the application of this formalism is the pulsating Harmonic Oscillator described by the periodically time-dependent Hamiltonian, which is a particular form of the Caldirola-Kanai Hamiltonian16 -19 H(q, p, t) =

1 p2 + m(t)ω02 q 2 2m(t) 2

(94)

where ω0 is the constant oscillator frequency and m(t) is a periodically time-varying mass such as18 m(t) = m0 e2µ sin νt

(95)

where µ is the strength of the pulsation and is the mass frequency. We shall find it convenient to introduce the canonical transformations r r m m0 Q=q P =p (96) m0 m The Hamiltonian becomes H(Q, P, t) =

1 m ˙ P Q + QP p2 + m0 ω02 Q2 + 2m0 2 m 2

(97)

From the form of Eq. (1), one can deduce the quantized forms of the unperturbed and perturbed Hamiltonians, respectively, such as 1 H0 = ~ω0 (a+ a + ) 2

(98)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

305

H1 (t) = −i~ν cos νt(a2 − a+2 )

(99)

where the annihilation operator a is defined as a= √

1 (β0 Q + iP ) 2~β0

(100)

By using Eqs. (5) and (14) and Eqs. (15) and (16) and choosing a frequency sequence such as ωk = 0, ν − 3ω0 , −(ν − 3ω0 ) we deduce HI = 0 fI (t) dH = −i~ν cos νt(e−2iω0 t a2 − e2iω0 t a+2 ) dt Integration of Eq. (102) gives fI (t) = γ0 (t)a2 + γ ∗ (t)a+2 H 0

where γ0 (t) = −

~ν e−2iω0 t (2ω0 cos νt + iν sin νt) ν 2 − 4ω02

(101) (102)

(103)

(104)

3.2.1. First order solutions i)Floquet operators By solving Eq. (4), using Eq. (103) and from Eq. (19), one can write + iµ (1a) e I eiω0 ta+ a e−iH0 t/~ =(1a) T (t)e−i(1a) Rt/~ U1 (t) = 1 − e−iω0 ta a H ~ (105) and deduce the first order ameliorated Floquet components such as (1a)

(1a)

with

T (t) = 1 −

R = H0

iµ γ1 (t)a2 + γ1∗ (t)a+2 ~

γ1 (t) = e2iω0 t γ0 (t)

(106) (107)

(108)

For a comparison with other works, we will calculate the position operator in the Heisenberg picture, by using the first order ameliorated evolution operator of Eq. (105). So one may write QH (t) =(1a) U1+ (t)Q(1a) U1 (t)

(109)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

306

To first order of µ we get QH (t) = Q(0) cos ω0 t + (

"

Q(0) ω0

P (0) 4µν sin ω0 t + × β0 4ω02 − ν 2

(110)

! # νt νt νt 2 cos − 1 sin ω0 t − ν sin cos cos ω0 t 2 2 2 2

" ! #) νt νt P (0) 2 νt ω0 2 sin + − 1 cos ω0 t − ν sin cos sin ω0 t β0 2 2 2 with Q(0) = Q and P (0) = P.

(111)

e If we take into account the initial condition H(0) = 0 in the integration of Eq. (102), this result is rather comparable with that obtained by Abdalla et al.18 which have employed the method of the perturbation theory. The comparison is also verified for PH (t) and aH (t). ii)Floquet states and uncertainty relation From Eqs. (6) and (8) and using Eqs. (106) and (107), we have calculated the first order Floquet states, " ( p iµ (1a) −i(n+1/2)ω0 t n(n − 1)γ1 (t)|n − 2i | ψn (t)i = e |ni − ~ #) (112) p ∗ + (n + 1)(n + 2)γ1 (t)|n + 2i

The expectation values of q, p, q 2 and p2 are hqi = 0

(113)

hpi = 0

(114)

h(1a) q 2 i =

~ (2n + 1){1 + µF1 (n, ν, t) + µ2 F2 (n, ν, t)} 2β

(115)

h(1a) p2 i =

~β (2n + 1){1 − µF1 (n, ν, t) + µ2 F2 (n, ν, t)} 2

(116)

where β(t) = m(t)ω0

(117)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

307

F1 (n, ν, t) = λ0 sin νt

(118)

λ1 2 2 (λ sin νt + cos2 νt) 2 0

(119)

ν 2ω0

(120)

4ν02 ω02 − 1

(121)

F2 (n, ν, t) = (n2 + n + 5) with λ0 = and λ1 =

Finally, calculation of the uncertainty relation in the Floquet-states (1a) |ψn (t)i gives (1a)

(∆q∆p) =

λ21 (1a) F (n, ν, t) = 4λ20

~ (2n + 1) 1 + µ2 2

2

n +n+1

λ20

2

(1a) F (n, ν, t)

2

12

(122)

2

sin νt+cos νt +4 cos νt

(123)

3.2.2. Second order solutions i)Floquet operators Solving the differential equation (21) and using Eqs. (101) and (103) we deduce Γ(t) = e−iµ

2

λ1

H0 t ~

(124)

Then, the second order solution (Eq. (20)) is obtained in the following form + iµ (2) e I eiω0 ta+ a e−i(1+µ2 λ1 )H0 t/~ =(2) T (t)e−i (2) Rt/~ U (t) = 1− e−iω0 ta a H ~ (125) Consequently, the second order Floquet operators are (2) (126) R = 1 + µ 2 λ 1 H0 (2)

T (t) = 1 −

iµ γ1 (t)a2 + γ1∗ (t)a+2 ~

(127)

December 29, 2009

9:37

WSPC - Proceedings Trim Size: 9in x 6in

recent

308

One observes that a µ-second order correction appears in (1a) R. Therefore, the ameliorated second order evolution operator is 2 iµ e 2 iH0 t/~ (2a) −iH0 t/~ e−i(1+µ λ1 )H0 t/~ − HI + µ A2 (t) e U (t) = 1 + e ~ (128) because SeI (t) = 0

then A2 (t) = − where

(129)

2 1 2 ∗ +2 f (t)H + γ (t)a + γ (t)a 0 0 0 0 2~2 f0 (t) = i~λ1

sin 2νt ν

(130)

(131)

Then, we deduce (2a)

R =(2) R (132) 2 iµ µ2 3 H0 (2a) 2 ∗ +2 2 T (t) = 1− γ1 (t)a +γ1 (t)a − 2 f0 (t)H0 +2|γ1 (t)| + ~ 2~ ~20 ω02 4 (133) +γ12 (t)a4 + γ1∗2 (t)a+4 A second order correction term in µ is also present in the operator

(2a)

T (t)

ii)Floquet states The eigenstates of the operator (2a)

(2a)

|φn i = e

R have the following form

−i(n+ 21 )(1+µ2 λ21 )ω0 t

|ni

(134)

From Eq. (6) and using Eq. (133), the construction of the ameliorated second order Floquet states leads to (2a) −i(n+ 12 )(1+µ2 λ21 )ω0 t n |ψn i = e C0n (t)|ni + C−2 (t)|n − 2i (135) n +C2n (t)|n + 2i + C−4 (t)|n − 4i + C4n (t)|n + 4i where the n-time-functions are µ2 1 n 2 2 C0 (t) = 1 − 2 n + ~ω0 f0 (t) + 2(n + n + 1)|γ1 (t)| 2~ 2

(136)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

309 n C−2 (t) = −

(137)

iµ p (n + 1)(n + 2)γ1∗ (t) ~

(138)

µ2 p n(n − 1)(n − 2)(n − 3)γ12 (t) 2~2

(139)

µ2 p (n + 1)(n + 2)(n + 3)(n + 4)γ1∗2 (t) 2~2

(140)

C2n (t) = − n C−4 (t) = −

C4n (t) = −

iµ p n(n − 1)γ1 (t) ~

iii)Heisenberg uncertainty relation The expectation values of q , p and q 2 in the |(2a) ψn (t)i states are (2a)

hqi =(2a) hpi = 0

(141)

n o ~ (2n+1) 1+µM1 (ν, t)+µ2 M2 (ν, t)+µ3 M3 (ν, t)+µ4 M4 (ν, t) 2β (142) n o ~β h(2a) p2 i = (2n+1) 1−µM1(ν, t)+µ2 M2 (ν, t)−µ3 M3 (ν, t)+µ4 M4 (ν, t) 2 (143) where 4i (144) M1 (ν, t) = Im(γ1 (t)) ~ h(2a) q 2 i =

M2 (ν, t) = 2(n2 + n + 4)

|γ1 (t)|2 ~2

(145)

i 2 2 2 M3 (ν, t) = 3 4(n + n + 5)|γ1 (t)| Imγ1 − (n + n + 1)~ω0 Re(γ1 )f0 (t) ~ (146) 2 Nm (n) 2 2 ω0 f0 (t) |γ1 (t)| − (2n + 1)2 M4 (ν, t) = (147) 4 ~ 4~ with Nm (n) =

1 2 (n + n + 2)(n2 + n + 52) + (n2 + n + 1)2 + 1 4

(148)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

310

The second order uncertainty product in Floquet-states (2a) |ψn (t)i is obtained in the following form 12 ~ (2a) (149) (∆q∆p) = (2n + 1) 1 + µ2 χ2 (n, ν, t) + µ4 χ4 (n, ν, t) 2 where 4 2 2 2 χ2 (n, ν, t) = 2 (n + n + 4)Re γ1 (t) − n(n + 1)Im γ1 (t) (150) ~ χ4 (n, ν, t) =

(ω0 |f0 (t)|) (2n+1)2 8~2

2

2 2 2 2 2 + 4 Nm (n)+2(n +n+4) |γ1 (t)| ~ (151)

3.2.3. Jump operators From Eqs. (11) and (12) we have determined the ladder operators in the following forms 2iµ ∗ + t i ∆ε ~ γ (t)a (152) a+ A(t) = e ~ 1 +

A (t) = e

−i ∆ε ~ t

2iµ a − γ1 (t)a ~ +

(153)

where γ1 (t) is given by Eq. (108) 3.3. Harmonic oscillator with a periodical frequency We also apply our approach to an Harmonic Oscillator with a periodically time-varying frequency Ω(t),5 ,1427 -.30 We consider the Hamiltonian of the form H(t) =

1 p2 + m0 Ω2 (t)q 2 2m0 2

(154)

By choosing a Mathieu frequency,5 ,14 ,2527 -30 or the so-called experiment’s Paul trap frequency,14 ,23 ,2729 -31 such as Ω2 (t) = ω02 (1 + µ cos 2ωt)

(155)

where µ and ω are the amplitude and the frequency of the oscillations, respectively. Then H(t) =

p2 1 µ + m0 ω02 q 2 + m0 ω02 cos 2ωtq 2 2m0 2 2

(156)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

311

The quantized form of the perturbation is H1 (t) =

~ω0 cos 2ωt(a2 + a+2 + 2aa+ + 1) 4

(157)

The RAM applied to the interaction picture form of H1 (t) (Eq. (5)) gives HI = 0

where

(158)

1 2 ∗ +2 + e HI = η0 (t)a + η0 (t)a + ~η2 (t) aa + 2 η0 (t) =

e−2iωt −i~ω0 −2iω0 t e2iωt e − 8 ω − ω0 ω + ω0

ω0 sin 2ωt 4ω Let us remark that our approach is very simple and the erator of this case is comparable to the one published by e I (t) in the standard picture we have, fact, if we write H 2 ∗ +2 e H1 (t) = η1 (t)a + η1 (t)a + ~η2 (t) aa+ +

(159)

(160)

η2 (t) =

(161) Hamiltonian opProfilo et al.5 In 1 2

(162)

with

η1 (t) = η1 (t)e2iω0 t

(163)

By differencing this operator, we have

where

e 1 (t) dH 1 = η˙1 (t)a2 + η˙1∗ (t)a+2 + ~η˙2 (t) aa+ + dt 2 h i ~ω0 ω ω cos(2ωt) + 2i sin(2ωt) η˙1 (t) = 2 ω 2 − ω02

(164)

(165)

By replacing it in the total Hamiltonian, we obtain 1 µ~ω0 ω b ω cos(2ωt) + 2iω0 sin(2ωt) a2 H(t) = ~Ω0 (t) a+ a + + 2 2 ω2 − ω2 0

(166)

+2 + ω cos(2ωt) − 2iω0 sin(2ωt) a

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

312

where 0

Ω (t) = ω0

µ 1 + cos 2ωt 2

(167)

which is a first order development of Ω(t) given in Eq. (155). We see that this Hamiltonian, where 2ω represents the driven frequency, resembles that appearing in the context of electromagnetic and acoustic parametric interactions, and is similar to that obtained by Profilo and al.5 using the invariant method in the framework of the Lie Group. 3.3.1. First order Floquet operators and Floquet states From Eq. (19) and with the initial condition (1) VI (t0 = 1, one has iµ (1a) e I eiH0 t/~ e−iH0 t/~ U1 (t) = 1 − e−iH0 t/~ H ~

(168)

(1a)

(169)

From the decomposition of Eq. (2), we determined the first order ameliorated Floquet components such as

(1a)

R = H0

T (t) = 1 −

iµ e H1 (t) ~

(170)

e 1 (t) is given by Eq. (162). where H We also remark that, to first order approximation, the operator (1a) R is equal to the simple Oscillator Hamiltonian, and that the operator (1a) T (t) is unitary and periodic with the same period as H1 (t). From Eqs. (6) and (8) and using Eqs. (162) and (170), the first order Floquet states are obtained as 1 (1a) −i(n+1/2)ω0 t |ni (171) | ψn (t)i = e 1 − iµη2 (t) n + 2 −

i p iµ hp n(n − 1)η1 (t)|n − 2i + (n + 1)(n + 2)η1∗ (t)|n + 2i ~

Then, the wave-Floquet function may be written in the form 1 (1a) ψn (q, t) = e−i(n+1/2)ω0 t 1 − iµη2 (t) n + ϕn (q) 2

i p iµ hp − n(n − 1)η1 (t)ϕn−2 (q) + (n + 1)(n + 2)η1∗ (t)ϕn+2 (q) ~

(172)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

313

where ϕn (q) =

β0 ~π

21

2 1 √ e−β0 q /2~ Hn n 2 n!

r

β0 q ~

!

(173)

is the wave function of the free oscillator, and Hn is the Hermite polynomials. The expectation values of q, p, q 2 and p2 are deduced, and one has hqin,n = hpin,n = 0 h(1a) q 2 in,n = h(1a) p2 in,n = where

(174)

n o ~ (2n + 1) 1 + µξ1 (t) + µ2 ξ2 (n, t) 2β0

n o ~β0 (2n + 1) 1 − µξ1 (t) + µ2 ξ2 (n, t) 2

ω02 ξ1 (t) = 2 cos(2ωt) 2 ω 2 − ω02

ξ2 (n, t) =

(175)

(176)

(177)

ω2 3 1 ω2 5 n2 +n+ 0 sin2 (2ωt)−2 02 n2 +n+ sin2 (2ωt) 2 ω 4 2 2 32 ω − ω0 (178)

+(n2 + n + 5)

ξ20 (n, t) =

ω2 0 32 ω 2 − ω02

n2 +n+

7 2

ω02 ω 2 − ω02

sin2 (2ωt)−2

ω02 2 1 sin2 (2ωt) n +n+ ω2 4 (179)

+(n2 + n + 5)

ω02 ω 2 − ω02

The first order uncertainty relation in Floquet-states (1a) |ψn (t)i is obtained in the form 21 ~ (1a) (180) ∆q∆p = (2n + 1) 1 + µ2 Fn (t) 2 n,n Fn (t) =

ω02

16 ω 2 − ω02

1 2 ω02 11 −2 n+ 3n + 3n + sin2 (2ωt) (181) 4 2 ω2 2

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

314

+

ω02 2 2 n + n + 1 + 4 sin (2ωt) (ω 2 − ω02 )

The Heisenberg uncertainty condition ∆q∆p ≥

~ 2

is satisfied for all time t.

3.3.2. Second order Floquet operators and Floquet states Solving the differential equation (21) and using Eqs. (22), (23, (158) and (159), we deduce Γ(t) = e

−iµ2

2H t ω0 0 16~(ω2 −ω2 ) 0

(182)

Then, the second order evolution operator Eq. (20) is obtained in the form 2 ω02 −i 1+µ iµ e 2 ) H0 t/~ (2) 16(ω2 −ω0 (183) U (t) = 1 − H1 (t) e ~

The corresponding Floquet operators are ω02 (2) H0 R = 1 + µ2 2 2 16(ω − ω0 ) (2)

T (t) =(1a) T (t)

(184) (185)

The determination of the ameliorated second order solution (Eq. (24)) neeI (t) and H e 2 (t) . We get cessitates the calculation of the operators Z I 2 + 0 2 0∗ +2 e ZI (t) = 2~ δ0 (t) 2a a + 1 + δ2 (t)a − δ2 (t)a (186) with

δ0 (t) =

δ20 (t) and

−iω03 sin(4ωt) 27 ω ω 2 − ω02

(187)

ω0 e4iωt 2 ω0 e−4iωt ω02 − + e−2iω0 t = 7 2 ω(ω + ω0 )(2ω + ω0 ) ω(ω − ω0 )(2ω − ω0 ) ω 2 − ω02 (188)

2 1 3 + 2 2 4 ∗2 +4 2 2 2 2 e a a+ HI (t) = η0 (t)a + η0 (t)a + ~ η2 (t) + 2|η0 (t)| + |η02 (t)|2 2 2 (189) ∗ +~η0 (t)η2 (t) 2a+ a + 3 a2 + ~ η0 (t)η2 (t) a+2 2a+ a + 3

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

315

The operator A2 (t) , given by Eq. (25), has the following form 1 e e2 (190) A2 (t) = 2 Z I (t) − HI (t) 2~ Therefore, the ameliorated second order evolution operator has the form 2 µ2 ω 0 −i 1+ H0 t/~ iµ e 2 2 (2a) −iH0 t/~ 2 iH0 t/~ 16(ω −ω0 ) − HI +µ A2 (t) e e U (t) = 1+e ~ (191) One may deduce the Floquet operators such as (2a)

R =(2) R

(192)

T (t) =(2) T (t) + µ2 A02 (t)

(193)

A02 (t) = e−iH0 t/~ A2 (t)eiH0 t/~

(194)

(2a)

with

Here the second order correction of has affected the operator (2) T (t). According to Eq. (6), the second order Floquet states are given by |(2a) ψn (t)i =(2a) T (t)|(2a) φn (t)i where |(2a) φn (t)i, the eigenstates of operator |(2a) φn (t)i = e

(2a)

(195)

R, are such as

2 µ2 ω 0 )ω0 t −i(n+ 21 )(1+ 16(ω2 −ω2 ) 0

|ni

(196)

The construction of the ameliorated second order Floquet states gives (2a) ε (ω)t n (2a) −i n ~ | ψn (t)i = e C0n (t)|ni + C−2 (t)|n − 2i + C2n (t)|n + 2i n +C−4 (t)|n − 4i + C4n (t)|n + 4i

(197)

where the second order quasi-energy is given by 1 µ2 ω02 (2a) εn (ω) = ~ω0 n + 1+ 2 16(ω 2 − ω02 )

(198)

The n-time-dependent coefficients are such as 3 1 C0n (t) = 1−iµ n+ η2 (t)+µ2 − 2 |η1 (t)|2 +(2n+1)δ0 (t)+(2n+1)2 δ1 (t) 2 2~ (199) h i p −iµ n (200) C−2 (t) = n(n − 1) η1 (t) + µ2 δ2 (t) + (2n − 1)δ3 (t) ~

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

316

C2n (t) =

p

(n + 1)(n + 2)

h i −iµ ∗ η1 (t) + µ2 − δ2∗ (t) + (2n + 3)δ3∗ (t) ~

n (t) = µ2 C−4

C4n (t) = µ2 with

p

p

(201)

n(n − 1)(n − 2)(n − 3)δ4 (t)

(202)

(n + 1)(n + 2)(n + 3)(n + 4)δ4∗ (t)

(203)

−1 2 2 2 |η1 (t)| + η2 (t) δ1 (t) = 4 ~2

(204)

δ2 (t) = δ20 (t)e2iω0 t

(205)

δ3 (t) = −

1 η1 (t)η2 (t) 2~

δ4 (t) =

−η12 (t) 2~2

(206)

(207)

The expectation values of q, p, q 2 and p2 in the |(2a) ψn (t)i states are such as (2a)

h

(2a) 2

q in,n

h(2a) p2 in,n

hqin,n =(2a) hpin,n = 0

(208)

~ n 2 n 3 n 4 n (2n + 1) 1 + µf1 (t) + µ f2+ (t) + µ f3+ (t) + µ f4+ (t) = 2β0 (209) ~β0 n n n = (2n + 1) 1 − µf1n (t) + µ2 f2− (t) + µ3 f3− (t) + µ4 f4− (t) 2 (210)

where f1n = n f2±

4Re(iη1 (t)) ~

(211)

2 |η1 |2 1 η22 + 2(2n + 1)2 δ1 ∓ 4Re(δ2 ) (212) = (2n + 2n + 7) 2 + n + ~ 2 2

±4(n2 + n + 3)Re(δ3 ) ∓

2i 2 (n + n + 1)η2 Im(iη1 ) ~

4i n f3± (t) = (2n + 1)2 η2 iδ0 ∓ 2Re(iδ3 ) + (n2 + n + 5)Im(η1∗ δ2 ) ~

(213)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

317

±

16 2 20 (n + n + 3)Re(iη1 δ4∗ ) + (2n2 + 2n + 3)Re(iη1 δ3∗ ) ~ ~ 3|η1 |2 4 ∓ (2n + 1)2 δ1 − Re(iη1 ) ~ 2~2

Im(δ2 ) Im(iη1 ) + iη2 − Im(δ3 ) ±4(n2 + n + 1) δ0 ~ 2

9 2 2 6|η1 |2 2 |η1 | ∓ Re (n + n + 3)δ3 − δ2 + 2(n2 + n + 5)|δ2|2 + 4 2 4~ ~ (214) 3|η1 |2 N0 (n)|δ3 |2 +(2n+1)2 −δ02 ±4Re(δ0 δ3 )− 2 δ1 ±4δ1 Re (n2 +n+3)δ3 −δ2 ~

n f4± (t) =

+(2n + 1)2 δ12

− 20(2n2 + 2n + 3)Re(δ2 δ3∗ ) ± (4n2 + n + 1)δ0 Im(2δ3 − δ2 )

∓16(n2 + n + 3)Re(δ2 δ4∗ ) + N1 (n, 6)|δ4 |2 ± 2N1 (n, 4)Re(δ3 δ4∗ ) where N0 (n) = 32n(n + 1) + 2(4n2 + 4n + 5)(n2 + n + 9) N1 (n, k)

(215)

(n + 4)! n! + + 2k (n2 + n + 3) n! (n − 4)!

(216)

Then, the second order uncertainty relation in Floquet-states (2a) |ψn (t)i has the form 21 ~ (2a) (217) (∆q∆p)n,n = (2n+1) 1+µ2 χn2 (ν, t)+µ3 χn3 (ν, t)+µ4 χn4 (ν, t) 2 where 16 2 η 2 (t) χn2 (ν, t) = (2n+1)2 4δ1 (t)+ 2 + 2 (2n2 +2n+7)|η1 (t)|2 − 2 (Re(iη1 (t))2 2 ~ ~ (218) 8i 2 (n + n + 5)Im(η1∗ (t)δ2 (t))+ (219) ~ 40 8 2 ∗ (2n + 2n + 3)Re(iη1 (t)δ3 (t)) + Re(iη1 (t)) 4Re(δ2 (t)) ~ ~

χn3 (ν, t) = 2i(2n + 1)2 η2 (t)δ0 (t) +

2i −4(n + n + 3)Re(δ3 (t)) + (n2 + n + 1)η2 (t)Im(iη1 (t)) ~ 2

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

318

3 1 |η12 | χn4 (ν, t) = (2n + 1)2 η22 (2n + 1)2 η22 + 3n2 + 3n + + (220) 16 2 ~2 8i 32i 2 2 Im(η1 )Re n + n + 3 δ3 − δ2 − 2δ0 − Im(η1 )Re(δ0 )Re(δ3 ) ~ ~

2 η22 Re2 (η1 ) 16i 2 2 Im(η1 )|η1 |2 +4(n + n + 1) (n + n + 1) 4 (|η1 | ) − + ~ ~2 ~3 2

2

h 4 8i − Re(η1 )η2 Re((n2 +n+3)δ3 −δ2 )+Re2 (δ2 )+ Im(η1 ) Re(δ0 )Im(δ2 −2δ3 ) ~ ~ +Im(δ0 )Im(δ2 )

i

+

4 2 (n + n − 3)2 (|η1 |2 )2 − 4(n2 + n + 5)Im2 (δ2 ) ~4

2

2

+ 2N0 (n) − 16(n + n + 3) Re2 (δ3 ) − 2N0 Im2 (δ3 ) + 2N1 (n, 6)|δ4 |2 16iIm(η1 ) 2 8(n + n + 3) Re(δ2 )Re(δ4 ) −40(2n + 2n + 3)Im(δ3 )Im(δ2 ) + ~ 2

+Im(δ2 )Im(δ4 )

− N1 (n, 4) Re(δ3 )Re(δ4 ) − Im(δ3 )Im(δ4 )

4i −24(2n + 2n + 1) Re(δ3 )Re(δ2 ) + Im(η1 )Im(δ0 )Im(δ3 ) ~ 2

3.4. Harmonic Oscillator with time-dependent mass and frequency The last system considered in this study is the time-dependent Harmonic Oscillator described by the periodically time-varying Hamiltonian such as H(t) =

p2 1 + m(t)Ω2 (t)q 2 2m(t) 2

(221)

where Ω(t) and m(t) are the time-varying frequency and mass respectively, given by Ω2 (t) = ω 2 (t)(1 + µf (t))

(222)

m(t) = m0 eµη(t)

(223)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

319

where µ is the strength of variation of Ω(t) and m(t) The equations η(t) and f (t) are two periodically time-varying functions satisfying η(t + T ) = η(t)

(224)

f (t + T ) = f (t) we introduce the canonical transformations Q=e

µη(t) 2

q

P =e

−µη(t) 2

p

(225)

The Hamiltonian becomes H(Q, P, T ) =

1 P Q + QP µ P2 + m0 ω02 Q2 + µη(t) ˙ + m0 ω02 f (t)Q2 (226) 2m0 2 2 2

we introduce the annihilation and creation operators a and a+ such as 1 (β0 Q + iP ) 2~β0

(227)

β0 = m 0 ω 0

(228)

a= √ where

By writing the Hamiltonian (Eq. (226)) in the quantized form of Eq. (1), one can deduce the unperturbed and perturbed Hamiltonians respectively as 1 H0 = ~ω0 (a+ a + ) (229) 2 i~ ~ω0 2 f (t)(a2 + a+2 + 2aa+ + 1) − η(t)(a ˙ − a+2 ) 4 2 choosing the functions H1 (t) =

η(t) = 2 sin νt

(230)

(231)

f (t) = cos 2ωt In the interaction picture form of H1 (t), and using the RAM technique, one may deduce the averaging and oscillating parts respectively as HI = 0 e I = ρ0 (t)a2 + ρ∗0 (t)a+2 + ~η2 (t) aa+ + 1 H 2

(232) (233)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

320

where ρ0 (t) =

−~ν 2

eiνt eiνt − ν − 2ω0 ν + 2ω0

η2 (t) =

i~ω0 − 16

e−2iωt e2iωt − ω − ω0 ω + ω0

e−2iω0 t

ω0 sin 2ωt 4ω

(234) (235)

3.4.1. First order solutions From Eq. (19), with the initial condition (1) VI (t0 = 1, we have iµ −iω0 ta+ a e −iω0 ta+ a iH0 t/~ (1a) (1a) HI e U (t) = 1 − e e = T (t)e−i ~

(1a)

Rt/~

(236)

The first order ameliorated Floquet operators are given by (1a)

(1a)

R = H0

(237)

T (t) = 1 − iµ ρ1 (t)a2 + ρ∗1 (t)a+2 + η2 (t) aa+ +

where

1 2

(238)

ρ0 (t) (239) ~ We notice that, to first order approximation, the operator (1a) R is equal to the unperturbed Hamiltonian, and the (1a) T (t) operator is unitary and periodic with the same period of H1 (t). From Eqs. (6) and (8) and using Eqs. (237) and (238), the first order Floquet states are such as ρ1 (t) = e2iω0 t

|

(1a)

ψn (t)i = e

−i(n+1/2)ω0 t

hp 1 |ni − iµ n(n−)ρ1 (t)|n − 2i 1 − iµη2 (t) n + 2

(240)

p + (n + 1)(n + 2)ρ∗1 (t)|n + 2i

i

The expectation values of Q, P , Q2 and P 2 are deduced and one gets hQin,n = hP in,n = 0

h(1a) Q2 in,n = h(1a) P 2 in,n =

n o ~ (2n + 1) 1 + µξ1 (t) + µ2 ξ2+ (t) 2β0

n o ~β0 (2n + 1) 1 − µξ1 (t) + µ2 ξ2− (t) 2

(241) (242) (243)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

321

where

ξ2± (t) =

1 n+ 2

2

ξ1 (t) = 4iIm ρ1 (t)

(244)

η22 (t) + 2(n2 + n + 5)|ρ1 (t)|2 ± 2(n2 + n + 1)η2 (t)Re(ρ1 (t)) (1a)

(245) |ψn (t)i is obtained

The first order uncertainty relation in Floquet-states in the following form 12 ~ (1a) (∆q∆p)n,n = (2n + 1) 1 + µ2 F1a (n, t) 2

(246)

2 1 η22 (t)+4(n2 +n+5)Re2 (ρ1 (t))−4(n2 +n+1)Im2 (ρ1 (t)) F1a (n, t) = 2 n+ 2 (247) The Heisenberg uncertainty condition ∆q∆p ≥ ~2 is satisfied for all time t. 3.4.2. Second order solutions As H I (t) = 0 and S I (t) = 0 , the differential equation (21) may be written in the following form i~

iµ2 dΓ Z I (t)Γ(t) = dt 2~

(248)

where Γ(t) = e−iµ

2

λmΩ

H0 t ~

(249)

and λmΩ =

4ν 2 ω02 + ν 2 − 4ω02 16(ω 2 − ω02 )

(250)

Then, the second order evolution operator (Eq. (20)) is obtained in the following form iµ −i H0 t e i H0 t −i(1+µ2 λmΩ ) H0 t (2) ~ ~ ~ e (251) HI e U (t) = 1 − e ~

The corresponding Floquet operators are (2)

R = (1 + µ2 λmΩ )H0

(252)

(2)

(253)

T (t) =(1a) T (t)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

322

Determination of the ameliorated second order solution (Eq. (24)) necessie 2 (t) such as tates the calculation of the operators ZeI (t), H I h i eI (t) = ~2 ϕ0 (t) 2a+ a + 1 + ϕ02 (t)a2 − ϕ02∗ (t)a+2 Z (254) with

(

ω02 sin 4ωt ν 2ν sin 2νt + + × 2 2 2 2 6 2 2 ν − 4ω0 2 ω(ω − ω0 ) 4(ν − 4ω0 )(ω 2 − ω02 ) (255) #) " (ν − 2ω)(νω + 2ω02 ) cos(ν + 2ω)t (ν + 2ω)(νω − 2ω02 ) cos(ν − 2ω)t − ν + 2ω ν − 2ω 2

ϕ0 (t) = −i~ ω0

ϕ02 (t)

~2 ω0 −2iω0 t = e 8ω

(

# " ω0 e4iνt 2ω ω0 e−4iνt ω0 − + 8 (ω + ω0 )(2ω + ω0 ) (ω − ω0 )(2ω − ω0 ) ω 2 − ω02

(256) +iν

"

(ν + 2(ω − ω0 ))ei(ν−2ω)t (ν − 2(ω + ω0 ))ei(ν+2ω)t − (ν − 2ω9 )(ν − 2(ω + ω0 )) (ν − 2ω9 )(ν + 2(ω − ω0 ))

(ν + 2(ω + ω0 ))e−i(ν−2ω)t (ν − 2(ω − ω0 ))e−i(ν+2ω)t + − (ν + 2ω9 )(ν − 2(ω − ω0 )) (ν + 2ω9 )(ν + 2(ω + ω0 ))

#)

and e 2 = ρ2 (t)a4 + ρ∗2 (t)a+4 + |ρ0 (t)|2 (a2 a+2 + a+2 a2 ) H I 0 0

(257)

The operator A2 (t) given by Eq. (25) becomes 1 e e2 A2 (t) = 2 Z (258) I (t) − HI (t) 2~ So, the ameliorated second order evolution operator is given by H t H0 t H0 t 2 iµ e 2 i ~0 (2a) + µ A (t) e e−i(1+µ λmΩ ) ~ (259) U (t) = 1 + e−i ~ − H I 2 ~ One may deduce the following Floquet operators (2a) (2a)

R =(2) R

(260)

T (t) =(2) T (t) + µ2 A02 (t)

(261)

with A02 (t) = e−i

H0 t ~

A2 (t)ei

H0 t ~

(262)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

323

then (2a)

1 µ2 T (t) = 1 − iµ ρ1 a2 + ρ∗1 a2 + η2 a+ a + + × 2 2

(263)

2 1 3 2ϕ0 + 1 2 2 ∗ +2 2 4 ∗2 +4 + a a+ + −2|ρ | +ϕ a −ϕ a −ρ a −ρ a a a+ 1 2 2 1 1 ~2 2 2 4

where ϕ2 (t) = e2iω0 t

ϕ02 (t) ~2

(264)

According to Eq. (6), the second order Floquet states are |(2a) ψn (t)i =(2a) T (t)|(2a) φn (t)i where |(2a) φn (t)i , the eigenstates of operator form 1

|(2a) φn (t)i = e−i(n+ 2 )(1+µ

2

(2a)

(265)

R , have the following

λmΩ )ω0 t

|ni

(266)

By substituting Eq. (263) into Eq. (265), the construction of the ameliorated second order Floquet states gives λmΩ t

n |(2a) ψn (t)i = e−i ~ {D0n (t)|ni + D−2 (t)|n − 2i + D2n (t)|n + 2i n n +D−4 (t)|n − 4i + D4 (t)|n + 4i}

D0n (t)

(267)

ϕ0 t 1 2 η2 (t)+µ −(n2 +n+1)|ρ1 (t)|2 +(2n+1) 2 (268) = 1−iµ n+ 2 2~

n D−2 (t) =

D2n (t) =

p

n ϕ2 (t) o n(n − 1) − iµρ1 (t) + µ2 2

n ϕ∗ (t) o (n + 1)(n + 2) − iµρ∗1 (t) − µ2 2 2

n D−4 (t) =

D4n (t) =

p

−µ p n(n − 1)(n − 2)(n − 3)ρ21 (t) 2

−µ p (n + 1)(n + 2)(n + 3)(n + 4)ρ∗2 1 (t) 2

(269)

(270) (271) (272)

The expectation values of q , p, q 2 and p2 in the |(2a) ψn (t) states are such as (2a)

hqin,n =(2a) hpin,n = 0

(273)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

324

h(2a) q 2 in,n =

~ n n n (2n+1){1+µg1n(t)+µ2 g2+ (t)+µ3 g3+ (t)+µ4 g4+ (t)} (274) 2β

h(2a) p2 in,n =

~β n n n (2n+1){1−µg1n(t)+µ2 g2− (t)+µ3 g3− (t)+µ4 g4− (t)} (275) 2

where β(t) = ω0 m(t)

(276)

g1n (t) = 4iIm(ρ1 (t))

(277)

1 2 n η ∓ 2Re(ϕ2 ) ± 2(n2 + n + 1)η2 Re(ρ1 ) (278) g2± (t) = 8|ρ1 |2 + n + 2 2 i h η 2 ϕ0 n g3± (t) = i(2n + 1)2 + 2i(n2 + n + 5) Im(ρ∗1 ϕ2 ) ± 2Im(η1 )|η1 |2 (279) 2~ ±2i(n2 + n + 1) n g4± (t) =

n+

1 2

2

ϕ Re(ρ ) η Im(ϕ ) 0 1 2 2 + ~2 2

(n2 + n + 5) |ϕ0 |2 N1 (n, 6) 2 2 |ρ1 |4 + + (n + n + 1) |ϕ2 |2 2 ~ 4 2 (280)

∓(n2 + n + 1)

ϕ0 Im(ϕ2 ) ± 2(3n2 + 3n + 7)Re(ϕ2 )Re2 (ρ1 ) ~2

±2(n2 + n + 5)Re(ϕ2 )Im2 (ρ1 ) ∓ 8(n2 + n + 3)Im(ϕ2 )Re(ρ1 )Im(ρ1 ) and N1 (n, k) =

(n + 4)! n! + + 2k (n2 + n + 3) n! (n − 4)!

(281)

Then, the second order uncertainty relation in Floquet-states (2a) |ψn (t)i has the following form 12 ~ (2a) (∆q∆p)n,n = (2n+1) 1+µ2 J2n (ν, t)+µ3 J3n (ν, t)+µ4 J4n (ν, t) (282) 2 where J2n (ν, t) = (2n + 1)2 J3n (ν, t) =

η22 + 16Re2 (ρ1 (t)) 2

i (2n + 1)2 η2 ϕ0 + 4i(n2 + n + 5)Re(ρ1 )Im(ϕ2 ) ~

(283) (284)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

325

−4i(n2 + n + 1)Im(ρ1 )[Re(ϕ2 ) + 4η2 Re(ρ1 )] 2 1 η24 |ϕ0 |2 2 2 + 4|ρ | η + n + + (n2 + n + 5)|ϕ2 |2 1 2 2~2 2 4 (285) N1 (n, 6) + 2(n2 + n + 1)2 + 64 Re4 (ρ1 ) − 4Re2 (ϕ2 ) + 2

J4n (ν, t) = (2n + 1)2

N1 (n, 6) 2 2 2 + 2(n + n + 1) − 32(n + n + 3) Im4 (ρ1 ) + 2

−(N1 (n, 6) − 4(n2 + n + 1)(n2 + n − 7))Re2 (ρ1 )Im2 (ρ1 ) +8(n2 + n + 1)

2ϕ0 Re(ρ1 )Im(ρ1 ) + η2 Im(ρ1 )Im(ϕ2 ) ~2

−4(n2 + n + 1)[(n2 + n + 1)η22 Re2 (ρ1 ) + 2η2 Re(ρ1 )Re(ϕ2 )] 4. Conclusion When the time variation of the hamiltonian of a quantum system is periodic, the Floquet theory is one of the most useful approaches which permits resolution of the Schr¨ odinger equation. Therefore, in this paper, we have treated the Floquet decomposition, solution of the evolution equation, by using the resonating averages method from the first order in µ to the second ameliorated order. The approach was applied to the forced harmonic oscillator, to the harmonic oscillator with a time-dependent mass and a constant frequency, to the harmonic oscillator with a constant mass and a timedependent frequency and to the harmonic oscillator with time-dependent mass and frequency. We have determined the approached forms of the evolution operator up to second ameliorated order in µ . Then, we have calculated the Floquet operators, Floquet states and verified that the uncertainty principle of Heisenberg is satisfied. Comparisons of our results, corresponding to each case of the above mentioned quantum systems, showed that they are identical to the results published by other authors which have used different methods,3 ,59 -,1218 This perfect concordance showed the efficiency of our approach and the validity of the method.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

326

References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31.

J.H. Shirley , Phys. Rev. B 138 (1965) 979 G. Lochak etMumm Thioun , le journal de physique 30 (1969) 482 H.P. Breuer and F. Petruccione, Phys. Rev. A 55 (1997) 3101 H.R. Lewis Jr., Phys.Rev.Lett. 27 (1967) 510; J.Math. Phys. 9 (1968) 1976 G. Profilo and G. Soliana ,Phys. Rev. A 44 (1991) 2057 C. F. Lo, Am. J. Phys. 59(1991) 254-258 ; IL Nuovo Cimento 105 (1990) 497-506 K.H. Yeon, K.K. Lee, Ch.I. Um, T.F. George and L.N. Pandey,Phys.Rev. A48 (1993) 2716 S. Pepore, P. Winotaia, T.Osotchanb and U. Robkob, Sc. Asia 32 (2006) 173-179 R.F. Fox and L.V. Vela-Arevalo, Phys. Rev. A 66 (2002) 053402 R. Lefebvre and A. Palma, J. Mol. Struct. (Theochem)390 (1997) 23-32 Neepa T. Maitra and KieronBurke,Chem.Phys. Lett. 359 (2002) 237-240 Prasanjit Samaland Manoj K. Harbola, Chem. Phys. Lett. 433 (2006) 204-210 Sara Cruz y Cruz and Bogdan Mielnik, Phys. Lett. A352(2006) 36-40 Francisco Delgado C. and Bogdan Mielnik,Phys. Lett. A249 (1998) 369-375 M.L. Glasser, L.M.Nieto and B.F. Samsonov J. Phys. A: Math. Gen. 36 (2003) L1-L7 E. Kanai, Prog. Theor. Phys. 3 (1948) 440 P.Caldirola,Nuovo cimento 18 (1941) 393 M.S. Abdalla, R.K.Colegrave and A. Khorsravi , IL Nuovo Cimento 93 B (1986) A. Geralico, G. Landolfi, G. Ruggeri and G. Soliani,arXiv:gr-qc/0308039 v2 31 Dec 2003 N . A. Lemos and C.P. Natividade, Nuovo Cimento 399 (1989) 211 S. Mandal,Opt. Comm. 240 (2004) 363-378 S. Mandal, Phys. Lett. A321 (2004) 308-318 W. Paul, Rev. Mod. Phys. 62 (1990)1947 Y. Achkar, S. Sayouri and A.L. Marrakchi, Phys.Chem. News 39 (2008) 19-24 A.Fedoul, A.L.Marrakchi,S.Sayouri and A.Chatwiti, Phys.Chem.News 24 (2005) 30-34 H. P. Breuer, K. Dietz and M. Holthaus, J. Phys. B: At.Mol. Opt. Phys. 24 (1991) 1343-1357 K.H. Yeon C. I. Um,and T. F. George, Phys. Rev.A 68 ( 2003) 052108-1-9 St.Weigert, J. Phys. A: Math. Gen 35 (2002) 4169 D.Schuch, Phys. Lett. A 338 (2005) 225-231 G. S. Agarwal and S. A. Kumar, Phys. Rev. Lett. 67 (1991) 3665 L. S.Brown, Phys. Rev. Lett. 66 (1991) 527

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

327

Renormalized solutions of nonlinear degenerated parabolic problems: Existence and uniqueness Y. Akdim, J. Bennouna, M. Mekkour, and M. Rhoudaf D´ epartement de Math´ ematiques Facult´ e des Sciences Dhar-Mahraz F` es, Morocco In this paper, we study the existence and uniqueness of renormalized solutions for the nonlinear degenerated parabolic problem ∂u − div(a(x, t, Du)) = f ∂t u(x, 0) = u0

in Ω × [0, T ],

in Ω

u = 0 in ∂Ω × [0, T ] where a : Ω × [0, T ] × RN → RN is a Carath´eodory function satisfying the coercivity condition, the general growth condition and only the large monotonicity. The second term f belongs to L1 (Q) and u0 ∈ L1 (Ω). Keywords: Weighted Sobolev spaces; Truncations; Renormalized solutions.

1. Introduction Let Ω be a bounded open set of RN , p be a real number such that 2 < p < ∞, Q = Ω × [0, T ] and w = {wi (x), 0 ≤ i ≤ N } be a vector of weight functions (i.e., every component wi (x) is a measurable function which is positive a.e. in Ω) satisfying some integrability conditions. The objective of this paper is to study the following problem, in the weighted Sobolev space, ∂u + Au = f in Ω × [0, T ], ∂t u = 0 on ∂Ω×]0, T [, u(x, 0) = u0

(1)

on Ω

where A = −div(a(x, t, Du)) and a : Ω × [0, T ] × RN → RN is a Carath´eodory function, satisfying the coercivity condition a(x, t, ξ).ξ ≥ α

N X i=1

wi |ξi |p ,

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

328

the general growth condition |ai (x, t, ξ)| ≤

1/p βwi (x)[k(x, t)

+

N X j=1

1/p0

wj

(x)|ξj |p−1 ],

and only the large monotonicity, [a(x, t, ξ) − a(x, t, η)](ξ − η) ≥ 0 for all (ξ, η) ∈ RN × RN , f belongs to L1 (Q). The difficulties that arise in problem 1 are due to the following facts: the data f belongs to L1 (Q) and a is monotone (but not necessarily strictly monotone). In the classical Sobolev space, existence of a weak solution has been established by L. Boccardo and T. Gallouet,2 and by D.Blanchard3 when a is strictly monotone. We consider here renormalized solutions, for reason, this definition allows us to prove the uniqueness of a renormalized solution for problem 1, in contrast to the framework of weak solution where this question is still open. The notion of renormalized solution was introduced by J. Diperna and P.L. Lions8 for the study of the Boltzmann equation. For the parabolic equation (1) the existence and uniqueness of a renormalized solution has been proved by D. Blanchard and F. Murat4 in the classical Sobolev space, they only supposed the large monotonicity. By L. Aharouch et al in1 in the weighted Sobolev space with f belongs to the dual space and a is strictly monotone. It is our purpose in this paper to generalize the result of 4 and prove the existence and uniqueness of renormalized solution for the problem (1.1) in the setting of the weighted Sobolev spaces. The proof uses techniques different from that given in [,34 ]. 2. Preliminaries and basic assumptions Let Ω be a bounded open set of RN , p be a real number such that 2 max(N, p) such that wir0 −p ∈ L1loc(Ω), −1 p−1

wi

∈ L1loc (Ω), for any 0 ≤ i ≤ N.

(2) (3)

W 1,p (Ω, w) is the space of all real-valued functions u ∈ Lp (Ω, w0 ) such that ∂u ∈ Lp (Ω, wi ). the derivatives in the sense of distributions ∂x i

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

329

Which is a Banach space under the norm N Z hZ i1/p X ∂u(x) p | kuk1,p,w = |u(x)|p w0 (x) dx + . | wi (x) dx ∂xi Ω i=1 Ω

(4)

The condition (3) implies that C0∞ (Ω) is a space of W 1,p (Ω, w) and consequently, we can introduce the subspace V = W01,p (Ω, w) of W 1,p (Ω, w) as the closure of C0∞ (Ω) with respect to the norm (4). Moreover, condition (3) implies that W 1,p (Ω, w) as well as W01,p (Ω, w) are reflexive Banach spaces. Assumption (H1) For 2 ≤ p < ∞, we assume that the expression N Z 1/p X ∂u(x) p | wi (x) dx | k|u|kV = ∂xi i=1 Ω

(5)

is a norm defined on V which equivalent to the norm (4), and there exist a weight function σ on Ω such that, σ ∈ L1 (Ω) and σ −1 ∈ L1 (Ω). We assume also the Hardy inequality, N Z X Z 1/q 1/p ∂u(x) p | ≤c , |u(x)|q σ dx | wi (x) dx ∂xi Ω i=1 Ω

(6)

holds for every u ∈ V with a constant c > 0 independent of u, and moreover, the imbedding W 1, p (Ω, w) ,→,→ Lp (Ω, σ),

(7)

expressed by the inequality (6) is compact. Note that (V, k|.|kV ) is a uniformly convex (and thus reflexive) Banach space. Assumption (H2) 1/p

|ai (x, t, ξ)| ≤ βwi

(x)[k(t, x) +

N X

1/p0

wj

j=1

(x)|ξj |p−1 ] f or 1 ≤ i ≤ N, (8)

[a(x, t, ξ) − a(x, t, η)](ξ − η) ≥ 0 for all (ξ, η) ∈ RN × RN , a(x, t, ξ).ξ ≥ α

N X i=1

wi |ξi |p ,

a(x, t, 0) = 0

(9) (10) (11)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

330 0

where k(x, t) is a positive function in Lp (Q), and α, β are strictly positive constants. We recall that, for k > 1 and s in R, the truncation Tk is defined as, ( s if s ≤ k Tk (s) = s k |s| if |s| > k. Rr And we define the function that will be used later ϕk (r) = 0 Tk (s)ds. 3. Main results Consider the problem u0 ∈ L1 (Ω),

f ∈ L1 (Q)

∂u − div(a(x, t, Du)) = f in Q ∂t u = 0 on ∂Ω×]0, T [, u(x, 0) = u0

(12)

on Ω.

Definition 3.1. Let f ∈ L1 (Q) and u0 ∈ L1 (Ω). A real-valued function u defined on Ω × [0, T ] is a renormalized solution of problem (12) if: u ∈ C([0, T ]; L1 (Ω));

(13)

Tk (u) ∈ Lp (0, T ; W01, p (Ω, w)) f or all (k ≥ 0);

(14)

∀ c ≥ 0, Tk+c (u)−Tk (u) → 0 strongly in Lp (0, T ; W01, p (Ω, w)) as k →+∞ (15) ∂S(u) − div (S 0 (u)a(x, t, Du)) + S 00 (u)a(x, t, Du)Du = f S 0 (u) in D0 (Q) ∂t (16) for all functions S ∈ C ∞ (R) such that S 0 has a compact support in R (i.e S 0 ∈ D(R)), u(t = 0) = u0 .

(17)

Remark 1. Equation (16) is formally obtained by pointwise multiplication of equation (12) by S 0 (u). Nevertheless, this process cannot be justified in general, so that weak solutions are not always renormalized solutions. In order terms, equation (16) is nothing but Z Z Z Z ∂ϕ 0 00 − S(u) + S (u)a(Du)Dϕ + S (u)a(Du)Duϕ = f S 0 (u)ϕ, ∂t Q Q Q Q

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

331

for all ϕ ∈ C0∞ (Q) and all S ∈ C ∞ (R) with S 0 ∈ C0∞ (R). This formally corresponds to using in problem (12) the test function ϕS(u), but once again, this is only formal. Indeed, if M is such that suppS 0 ⊂ [−M, M ], the following identifications are made in (16): • S(u) belongs to L∞ (Q) since S is a bounded function. • S 0 (u)a(x, t, Du) identifies with S 0 (u)a(x, t, DTM (u)) a.e in Q. Since TM (u) ≤ M a.e in Q and S 0 (u) ∈ L∞ (Q), we obtain from (8) and (14) that S 0 (u)a(x, t, DTM (u)) ∈

N Y

0

Lp (Q, wi∗ )

i=1

• S 00 (u)a(x, t, Du)Du identifies with S 00 (u)a(DTM (u))DTM (u) and S 00 (u)a(DTM (u))DTM (u) ∈ L1 (Q). • S 0 (u)f belongs to L1 (Q) . The above considerations show that equation (16) holds in D 0 (Q) and ∂S(u) ∂t 0 0 belongs to Lp (0, T ; W −1, p (Ω, wi∗ )) + L1 (Q). It follows that u belongs to C 0 (0, T ; L1 (Ω)) so that the initial condition (17) makes sense. Theorem 3.1. Let f ∈ L1 (Q) and u0 ∈ L1 (Ω). Assume that (H1) and (H2), there exists a unique renormalized solution u of problem (12) . Proof of the existence result Step 1: The approximate problem Consider the approximate problem: un ∈ Lp (0, T ; W01, p (Ω, w))

∂un − div(a(x, t, Dun )) = fn in D0 (Q), ∂t un (t = 0) = u0n ,

(18)

with (fn ) be a sequence of smooth functions such that fn = Tn (f ) and fn → f in L1 (Q), and (u0n ) be a sequence such that u0n = Tn (u0 ) and u0n → u0 in L1 (Ω). fn and u0n belong, respectively to L∞ (Q) and L∞ (Ω). 0 0 Let X = Lp (0, T ; W01, p (Ω, w)), X ∗ = Lp (0, T ; W −1,p (Ω, w∗ )), H = L2 (Ω, σ) and Wp1 (0, T, V, H) = {v ∈ X : v 0 ∈ X ∗ }. The operator A is bounded, demicontinuous, pseudomonone with respect to D(L) where D(L) = {v ∈ X; v 0 ∈ X ∗ , v(0) = 0} and strongly coercive. So all conditions of theorem 5 in5 are met. Therefore, there exists a solution

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

332

un ∈ D(L) of the evolution equation (18) for any fn ∈ X ∗ . By the definition of D(L) and see,1 ,10 we obtain D(L) ⊆ Wp1 (0, T, V, H) ⊆ C([0, T ], H). Which implies that un ∈ C([0, T ], H). Step 2: Some estimations about the truncated sequence of solutions. Using in (18) the test function Tk (un )χ(0,τ ) , we get, for every τ ∈ [0, T ] (Since |ϕk (r)| ≤ k |r|). Z Z ∂un ϕk (un (τ ))dx − k ku0n kL1 (Ω) . Tk (un )dx ≥ Ω Q ∂t Then we deduce that, Rτ R R ϕ (u (τ ))dx + 0 Ω a(x, t, Dun )DTk (un )dxdt Ω k n ≤ k(ku0n kL1 (Ω) + kfn kL1 (Q) ) ≤ ck. Thanks to (10) and by the fact that ϕk (un (τ )) ≥ 0, we deduce that, N R P k (un ) p α Q wi (x) ∂T∂x dxdt i i=1

(19)

≤ ck, ∀k ≥ 1.

Then, Tk (un ) is bounded in Lp (0, T ; W01, p (Ω, w)), and Tk (un ) * vk in Lp (0, T ; W01, p (Ω, w)). Using the compact imbedding (3.3) we get, Tk (un ) → vk strongly in Lp (Q, σ) and a.e in Q. Let k > 0 large enough and BR be a ball of Ω, we have, RT R k meas({|un | > k} ∩ BR × [0, T ]) = 0 {|un |>k}∩BR |Tk (un )| dxdt RT R ≤ 0 BR |Tk (un )| dxdt ≤ T cR So, we have

! p1 ∂Tk (un ) p 1 dxdt wi (x) ≤ ck p ∂x i i=1

Z X N Q

lim (meas({|un | > k} ∩ BR × [0, T ])) = 0.

k→+∞

Consider now a function non decreasing gk ∈ C 2 (R) such that gk (s) = s f or |s| ≤ k2 and gk (s) = k f or |s| ≥ k Multiplying the approximate equation by gk0 (un ), we get ∂gk (un ) − div(a(x, t, Dun )gk0 (un )) + a(x, t, Dun )gk00 (un ) = fn gk0 (un ) ∂t

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

333

in the sense of distributions. This implies, thanks to the fact gk0 has compact (un ) support, that gk (un ) is bounded in X, while it’s time derivative ∂gk∂t is ∗ 1 bounded in X + L (Q). Hence see Lemma 3.8 in1 allows us to conclude that gk (un ) is compact in Lploc (Q, σ). Thus, for a subsequence, it also converges in measure and almost every where in Q (since we have, for every λ > 0, ) meas({|un − um | > λ} ∩ BR × [0, T ]) ≤ meas({|un | > k} ∩ BR × [0, T ]) +meas({|um | > k} ∩ BR × [0, T ]) + meas({|gk (un ) − gk (um )| > λ}) Let ε > 0, then, there exist k(ε) > 0 such that, meas({|un − um | > λ} ∩ BR × [0, T ]) ≤ ε f or all n, m ≥ n0 (k(ε), λ, R). This proves that (un ) is a Cauchy sequence in measure in BR × [0, T ], thus converges almost everywhere to some measurable function u. Then for a subsequence denoted again un , we can deduce from (19) that, Tk (un ) * Tk (u) weakly in Lp (0, T ; W01, p (Ω, w))

(20)

and then, the compact imbedding (3.3) gives, Tk (un ) → Tk (u) strongly in Lp (Q, σ) and a.e in Q. Which implies, by using (8), for all k > 0 that there exists a function N Q 0 hk ∈ Lp (Q, wi∗ ), such that i=1

a(x, t, DTk (un )) * hk weakly in

N Y

0

Lp (Q, wi∗ )

(21)

i=1

We now establish that u belongs to L∞ (0, T ; L1 (Ω)). Using (19), (20) and passing to the limit-inf as n tends to +∞ and we obtain Z ϕk (u)(t)dx ≤ k[kf kL1 (Ω) + ku0 kL1 (Ω) ] Ω

By the definition of ϕk , we deduce from the above inequality that Z 3k 2 meas(Ω) + k[kf kL1 (Ω) + ku0 kL1 (Ω) ] k |u(x, t)| dx ≤ 2 Ω for almost any t in (0,T), which gives that u belong to L∞ (0, T ; L1(Ω)) . Step 3: In this step we identify the weak hk in (21) and we prove the

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

334

weak-L1 convergence of the ”truncated” energy a(x, t, DTk (un ))DTk (un ) as n tends to + ∞. We can prove easily the limits Z a(Dun )Dun dxdt = 0. (22) lim lim sup m→+∞ n→+∞

{m≤|un |≤m+1}

We define for all µ ≥ 0 and all (x, t) ∈ Q, Z t vµ = µ v˜(x, s) exp(µ(s − t))ds where v˜(x, s) = v(x, s)χ(0,T ) (s). ∞

See proposition 3.1-3.3 in.1

Lemma 3.1. Let k ≥ 0 be fixed. Let (Tk (u))µ the mollification of Tk (u). Let S be an increasing C ∞ (R)-function such that S(r) = r f or |r| ≤ k and supp S 0 is compact. Then, Z T Z t ∂S(un ) lim lim , (Tk (un ) − (Tk (u))µ ) dtds ≥ 0, (23) µ→+∞ n→+∞ 0 ∂t 0 where h., .i denotes the duality pairing 1, p 1 −1,p0 ∗ ∞ L (Ω) + W (Ω, w ) and L (Ω) ∩ W0 (Ω, w).

between

Proof of Lemma 3.1. Let k ≥ 0 be fixed. Let us first recall that since supp S 0 is compact, we have ∂S(un ) ∈ L1 (Q) + X ∗ . ∂t Moreover, since S is increasing and S(r) = r f or |r| ≤ k, S(un ) ∈ X ∩ L∞ (Q), S(u) ∈ X ∩ L∞ (Q) and

Tk (S(un )) = Tk (un ) and Tk (S(u)) = Tk (u) a.e. in Q. As a consequence, (Tk (S(u)))µ = (Tk (u))µ a.e. in Q for any µ > 0. It follows that under the notation vn = S(un ) and v = S(u), we have Z T Z t ∂S(un ) , (Tk (un ) − (Tk (u))µ ) dtds 0 Z 0 Z ∂t T t ∂(vn ) = , (Tk (vn ) − (Tk (v))µ ) dtds ∂t 0 0 = − +

Z

Z

Z

T 0 T

0 T 0

∂(vn − (Tk (v))µ ) , (vn − (Tk (v))µ ) dtds ∂t 0 Z t ∂(vn ) , (vn − Tk (vn )) dtds ∂t 0 Z tZ ∂(Tk (v))µ (vn − (Tk (v))µ )dxdtds, ∂t 0 Ω Z t

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

335

Z T |(vn − (Tk (v))µ )|2 (t = 0)dx |(vn − (Tk (v))µ )| dxdt − 2 Ω 0 Ω Z Z Z T 1 T |vn − (Tk (vn ))|2 dxdt + |(vn − (Tk (vn )))|2 (t = 0)dx − 2 0 Ω 2 Ω Z T Z tZ ∂(Tk (v))µ + (vn − (Tk (v))µ )dxdtds, ∂t 0 0 Ω (24) Rr since 0 (s − Tk (s))ds = 12 |r − Tk (r)|2 . To pass to the limit in (24) as n tends to +∞, we first observe that since S is bounded, the definition of vn and v together with convergence of un a.e in Q imply that 1 = 2

Z

T

Z

2

vn → v strongly in L2 (Q) and in L∞ weak − ∗. Moreover, the initial condition for un gives vn (t = 0) = S(un )(t = 0) = S(un0 ) a.e in Q, so that the strong convergence of u0n to u0 in L1 (Ω) implies that vn (t = 0) → S(u0 ) in L2 (Ω). Passing to the limit as n tends to +∞ in (24) is an easy task and leads to Z T Z t ∂S(un ) , (Tk (un ) − (Tk (u))µ ) dtds lim n→+∞ 0 ∂t 0 Z Z Z T 1 T |(v − (Tk (v))µ )|2 dxdt − |(S(u0 ) − (Tk (v))µ )|2 (t = 0)dx = 2 0 Ω 2 Ω Z Z Z T 1 T |v − (Tk (v))|2 dxdt + |S(u0 ) − Tk (S(u0 ))|2 dx − 2 0 Ω 2 Ω Z T Z tZ ∂(Tk (v))µ + (v − (Tk (v))µ )dxdtds, ∂t 0 0 Ω (25) for any µ > 0. In order to pass to the limit-inf as µ tends to infinity in (25), we now use the definition of (Tk (u))µ , in terms of Tk (v), rewrite as ∂(Tk (v))µ + µ((Tk (v))µ − Tk (v)) = 0 in D0 (Q), ∂t

(26)

(Tk (v))µ (t = 0) = v0µ in Ω,

(27)

(Tk (v))µ → Tk (v) in L2 (Q)

(28)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

336

(Tk (v))µ (t = 0) → Tk (S(u0 )) in L2 (Q)

(29)

as µ tends to +∞, since again Tk (S(u0 )) = Tk (u0 ) a.e in Ω. In view of (26), (28), (29) passing to the limit-inf as µ tends to +∞ in (25) leads to Z T Z t ∂S(un ) lim inf lim , (Tk (un ) − (Tk (u))µ ) dtds µ→+∞ n→+∞ 0 ∂t 0 (30) Z T Z tZ (Tk (v) − (Tk (v))µ )(v − (Tk (v))µ )dxdtds. = lim inf µ µ→+∞

0

0

Ω

The proof of lemma 3.1 then reduces to show that the right-hand side of (30) is nonnegative. To this end we just remark that (Tk (v) − (Tk (v))µ )(v − (Tk (v))µ ) ≥ 0 a.e in Q. We prove the following lemma, which is the key point in the monotonicity arguments that will be developed in undo lemma . Lemma 3.2. The subsequence of un satisfies for any k ≥ 0 Z T Z tZ Z T Z tZ lim sup a(DTk (un ))DTk (un )dxdtds ≤ hk DTk (u)dxdtds, n→+∞

0

0

Ω

0

0

Ω

(31)

where hk is defined in (21). In the following we adapt the above-mentioned method to problem (12) and we first introduce a sequence of increasing C ∞ (R)-functions Sm such that Sm (r) = r if |r| ≤ m, 0 suppSm ⊂ [−(m + 1), m + 1], 00 kSm kL∞ ≤ 1, f or any m ≥ 1. 0 Pointwise multiplication of equation (18) by Sm (un ) (which is licit) leads to ∂Sm (un ) 0 − div (Sm (un )a(x, t, Dun )) ∂t 00 +Sm (un )a(x, t, Dun )Dun = fn S 0 (un )

in D0 (Q)

(32)

which is nothing but the renormalized of formulation (16) for un with (u) S = Sm . Remark that (32) implies that ∂Sm ∈ L1 (Q) + X ∗ . ∂t

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

337

The use of the test function Wµn = Tk (un ) − (Tk (u))µ , k ≥ 0 in equation.(32) we obtain upon integration over (0,t) and then over (0,T): Z T Z t Z T Z tZ ∂Sm (un ) 0 , Wµn dtds + Sm (un )a(Dun )DWµn ∂t 0 0 0 0 Ω Z T Z tZ Z T Z tZ 00 n 0 + Sm (un )a(Dun )Dun Wµ dxdtds = f n Sm (un )Wµn . 0

0

Ω

0

0

Ω

(33) We pass to the limit in (33) as n tends to +∞, µ tends to +∞ and then m tends to +∞, the real number k ≥ 0 being fixed. In order to perform this task we prove below the following results for fixed k ≥ 0 : Z T Z t ∂Sm (un ) , Wµn dtds ≥ 0, f or any m ≥ k, (34) lim inf lim µ→+∞ n→+∞ 0 ∂t 0 Z Z Z T t 00 n Sm (un )a(Dun )Dun Wµ dxdtds = 0, m ≥ 1 lim lim sup lim sup m→+∞ µ→+∞ n→+∞ 0 0 Ω

(35)

lim

lim

µ→+∞ n→+∞

Z

T 0

Z tZ 0

Ω

0 f n Sm (un )Wµn dxdtds = 0 f or any m ≥ 1 (36)

Proof of (34). The function Sm belongs to C ∞ (R) and is increasing. We 0 have for m ≥ k, Sm (r) = r f or |r| ≤ k while suppSm is compact. n In view of the definition of Wµ , lemma 3.1 applies with S = Sm for fixed m ≥ k. As a consequence (34) holds true. Proof of (35) In order to avoid repetitions in the proofs of (36), let us summarize the properties of Wµn . For fixed µ > 0 Wµn * Tk (u) − (Tk (u))µ weakly in Lp (0, T ; W01, p (Ω, w)), as n → +∞

n

W µ

L∞ (Q)

≤ 2k, f or any n > 0 and f or any µ > 0

we deduce that for fixed µ > 0 Wµn → Tk (u) − (Tk (u))µ a.e in Q and in L∞ (Q)weak − ∗, as n → +∞ 00 one has suppSm ⊂ [−(m + 1), −m] ∪ [m, m + 1] for any m ≥ 1. As a consequence R R R T t 00 (un )a(Dun )Dun Wµn 0 0 Ω Sm

R 00 ≤ T kSm (un )kL∞ Wµn L∞ {m≤|un |≤m+1} a(Dun )Dun

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

338

for any m ≥ 1, any µ > 0 and any n ≥ 1 it possible to obtain Z Z Z T t 00 n lim sup lim sup Sm (un )a(Dun )Dun Wµ dxdtds µ→+∞ n→+∞ 0 0 Ω ≤ C lim sup n→+∞

Z

a(Dun )Dun dxdt,

{m≤|un |≤m+1}

for any m ≥ 1, where C is a constant independent of m. Appealing now to (22) it possible to pass the limit as m tends to +∞ to establish (35). Proof of (36) Lebesque’s convergence theorem implies that for any µ > 0 and any m ≥ 1 RT RtR f S 0 (u )Wµn dxdtds lim 0 Ω n m n n→+∞ 0 RT RtR 0 = 0 0 Ω f Sm (u)(Tk (u) − (Tk (u)µ ))dxdtds

Now, for fixed m ≥ 1, using1 makes it possible to pass to the limit as µ → +∞ in the above equality to obtain (36). We now turn back to the proof of lemma 3.2. Due to (34), (35) and (36), we are in a position to pass the limit-sup when n tends to +∞, then to the limit-sup when µ tends +∞ and then to the limit as m tends to +∞ in (33). Using the definition of Wµn we obtain that for any k ≥ 0 RT RtR 0 lim lim sup lim sup 0 0 Ω Sm (un )a(Dun ) m→+∞ µ→+∞ n→+∞

(DTk (un ) − D(Tk (u))µ )dxdtds ≤ 0.

Since for k ≤ n and k ≤ m, the above inequality implies that for k ≤ m Z T Z tZ lim sup a(Dun )DTk (un )dxdtds n→+∞

≤

0

0

Ω

lim lim sup lim sup

m→+∞ µ→+∞ n→+∞

Z

T

0

Z tZ 0

Ω

0 Sm (un )a(Dun )D(Tk (u))µ dxdtds

(37) The right-hand side of (37) is computed as follows. We have for n ≥ m + 1: 0 0 Sm (un )a(Dun ) = Sm (un )a(DTm+1 (un )) a.e in Q.

Due to the weak convergence of a(DTm+1 (un )) it follows that for fixed m≥1 0 Sm (un )a(Dun ) * S 0 m(un )hm+1 weakly in

N Y

i=1

0

Lp (Q, wi∗ )

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

339

when n tends to +∞. The strong convergence of (Tk (u))µ to Tk (u) in Lp (0, T ; W01, p (Ω, w)) as µ tends to +∞, then we have Z T Z tZ 0 Sm (un )a(Dun )D(Tk (u))µ lim lim µ→+∞ n→+∞ 0 0 Ω (38) Z T Z tZ =

0

0

Ω

0 Sm (un )hm+1 DTk (u),

as soon as k ≤ m, S 0 m(r) = 1 f or |r| ≤ m. Now for k ≤ m we have, a(DTm+1 (un ))χ{|un |
hm+1 χ{|un |
(40)

Using (37), (38), (40) we obtain (31) and the proof of lemma 3.2 is complete. Let k ≥ 0 be fixed. Using (9) we have Z T Z tZ lim [a(DTk (un )) − a(DTk (u))][DTk (un ) − DTk (u)]dxdtds ≥ 0. n→+∞

0

0

Ω

(41) By lemma 3.2, weak convergence of Tk (un ) and a(DTk (un )) we pass to the limit-sup as n → +∞ in (41) and obtain the result : lim

n→+∞

Z

T 0

Z tZ 0

Ω

[a(DTk (un )) − a(DTk (u))][DTk (un ) − DTk (u)]dxdtds = 0. (42)

Lemma 3.3. For fixed k ≥ 0, we have hk = a(DTk (u)) a.e in Q, a(DTk (un ))DT (un ) * a(DTk (u))DTk (u) weakly in L1 (Q).

(43) (44)

Proof of Lemma 3.3. The proof is standard, from weak convergence of (Tk (un )) and a(DTk (un )), (42) and (9) we deduce the result. Step 4: Convergence of (un ) in C([0, T ], L1 (Ω)) Proposition 3.1. The sequence (un ) is a Cauchy sequence in C([0, T ], L1 (Ω)).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

340

Moreover, u C([0, T ], L1 (Ω)).

∈

C([0, T ], L1(Ω)) and un

converges to u in

The proof is similar of Proposition 7.3.1 in.9 Step 5: In this step we prove that u satisfies (15). Using in (18) the admissible test function Tk+c (un ) − Tk (un ) leads to Z Z [ϕk+c (un )(T ) − ϕk (un )(T )]dx + a(Dun )[DTk+c (un ) − DTk (un )]dxdt Ω Q Z Z fn [Tk+c (un ) − Tk (un )]dxdt + [ϕk+c (u0n ) − ϕk (u0n )]dx. = Ω

Q

(45) In (45) and in what follows the (t, x) dependence on a(x, t, d) is dropped for convenience. Since the function ϕk+c (r) − ϕk (r) is positive, using the convergence of fn to f in L1 (Q), the convergence of u0n to u0 in L1 (Ω), the fact that |Tk+c (r) − Tk (r)| ≤ C, the linear growth of ϕk at infinity, and (3.1), we deduce from (45)that: Z

lim sup a(Dun )[DTk+c (un ) − DTk (un )]dxdt n→∞ Q Z Z ≤ f [Tk+c (u) − Tk (u)]dxdt + [ϕk+c (u0 ) − ϕk (u0 )]dx. Q

(46)

Ω

Since DTk+c (un ) − DTk (un ) = 0 when (t, x) does not belong to the set {(t, x); k ≤ |un (t, x)| ≤ k + c}, we have almost pointwise in Q a(Dun )[DTk+c (un ) − DTk (un )]

= a(DTk+c (un ) − DTk (un ))[DTk+c (un ) − DTk (un )]

(47)

Since, by (20), Tk+c (un ) − Tk (un ) weakly converge to Tk+c (u) − Tk (u) in X as n tends to ∞, we obtain Z X N ∂[Tk+c (u) − Tk (u)] p wi (x)dxdt α lim sup ∂xi n→∞ Q i=1 (48) Z Z ≤ f [Tk+c (u) − Tk (u)]dxdt + [ϕk+c (u0 ) − ϕk (u0 )]dx. Q

Ω

When k tends to infinity, the right-hand side (48) tends to zero: indeed Tk+c (u) − Tk (u) is dominated by C and tends almost everywhere to zero, while Z Z [ϕk+c (u0 ) − ϕk (u0 )] ≤ C |u0 (x)| dx. Ω

{|u0 (x)|>k}

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

341

This proves that u satisfies (15). Step 6: Passing to the limits. We now return to the proof of the existence result of theorem (3.1). We notice that the function u defined in Lemma 20 meets the regularity requirements (15), (13) and (16) of Definition 3.1. The initial condition is easily recovered by passing to the limit in un (t = 0) = u0n , and using (3.1). It thus only remains to prove that u satisfies equan in tion (16). Since un belongs to Lp (0, T ; W01, p (Ω, w)) with ∂u ∂t p0 −1,p0 ∗ ∞ 0 ∞ L (0, T ; W (Ω, w )) , we have for any S ∈ C (R) with S ∈ C0 (R) ∂un ∂S(un ) = S 0 (un ) in D0 (Q). ∂t ∂t Using this fact in equation (18) leads to: ∂S(un ) − div[S 0 (un )a(Dun )] + S 00 (un )a(Dun )Dun = f S 0 (un ) in D0 (Q). ∂t (49) To pass to the limit in (49) as n tends to +∞, let M be such that supp S 0 ⊂ [−M, M ], so that S 0 (un )a(Dun ) = S 0 (un )a(DTM (un )). The pointwise convergence of un to u as n tends to +∞ and the bounded character of S 0 permit us to conclude that S 0 (un )a(Dun ) * S 0 (u)a(DTM (u)) in

N Y

0

Lp (Q, wi∗ ).

(50)

i=1

as n tends to +∞. S 0 (u)a(DTM (u)) has been denoted by S 0 (u)a(Du) in equation (16). As far as the ’energy’ term S 00 (un )a(Dun )Dun = S 00 (un )a(DTM (un ))DTM (un ) is concerned, remark that Z [a(DTM (un )) − a(DTM (u))][DTM (un ) − DTM (u)]dxdt = 0. lim n→+∞

Q

Due to the monotonicity (9), the above equality shows that [a(DTM (un ))−a(DTM (u))][DTM (un )−DTM (u)] → 0 strongly in L1 (Q) (51)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

342

as n tends to +∞. Again using (51), (20) and (44) then implies that a(DTM (un ))DTM (un ) * a(DTM (u))DTM (u) weakly in L1 (Q)

(52)

as n tends to +∞. Appealing now to the pointwise convergence of S 00 (un ) to S 00 (u) and since S 00 is bounded, a straightforward application of Egoroff’s theorem leads to S 00 (un )a(DTM (un ))DTM (un ) * S 00 (u)a(DTM (u))DTM (u) weakly in L1 (Q),

(53)

as n tends to +∞. Finally, the pointwise convergence (3.1) and the L1 (Q)−strong convergence of fn to f yield S 0 (un )fn → S 0 (u)f strongly in L1 (Q),

(54)

as n tends to +∞. The L∞ −weak* convergence of S(un ) to S(u) together with (49), (50), (53) and (54) proves that (16) is satisfied in the sense of distributions. This completes the proof of the existence result of Theorem 3.1. Proof of uniqueness result We now turn to the proof of the uniqueness result of Theorem 3.1. In fact, we will prove a comparison result for renormalised solutions. Lemma 3.4. Assume that u01 and u02 lie in L1 (Ω), that f1 and f2 lie in L1 (Q) and that they satisfy: u01 ≤ u02 ,

(55)

f1 ≤ f 2 .

(56)

Then if u1 and u2 are two renormalised solutions of problem (18) respectively for the data (f1 , u01 ) and (f2 , u02 ), we have u1 ≤ u2 almost everywhere in Q.

(57)

Remark 2. Uniqueness of a solution of (13)-(16) is a direct application of Lemma 3.4.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

343

References 1. L.Aharouch, E.Azroul and M.Rhoudaf, Strongly nonlinear variational parabolic problems in weighted sobolev spaces. The Australian journal of Mathematical Analysis and Applications, Vol.5, Issue 2, Article 13, pp. 125,2008 . 2. L. Boccardo and T.Gallouet. On some nonlinear elliptic and parabolic equations involving measure data.J.Funct.Anal.87(1989),149-69. 3. D. Blanchard. Truncations and monotonocity methods for parabolic equations. Nonlinear Anal.21 (1993),725-43. 4. D. Blanchard, F.Murat. Renormalized solutions of nonlinear parabolic problems with L1 data: existence and uniqueness. Proceedings of the Royal Society of Edinburgh, 127A, 1137-1152, 1997. 5. J.Berkovits and V.Mustonen. Topological degree for perturbations of linear maximal monotone mappings and applications to a class of parabolic problem . Rendiconti di Matematica, Serie VII, volume 12, Roma (1992),597-621. 6. P. Drabek, A. Kufner and F. Nicolosi, Non linear elliptic equations, singular and degenerated cases, University of West Bohemia, (1996). 7. J. L. Lions quelques m´ethodes de r´esolution des probl`eme aux limites non lineaires, Dundo, Paris (1969). 8. P. L. Lions and R.J.Diperna, On the cauchy problem for Boltzman equations: global existence and weak stability. Ann.of Math.(2)130 (1989),321-366. 9. M.Rhoudaf, Etude de certains problemes elliptiques et paraboliques non lineaires avec second membre L1 ou dans le dual , Th`ese de doctorat d’ etat, Facult´e des Sciences Dhar-Mhraz, F`es, Maroc, (2006) . 10. E.Zeidler , Nonlinear Functional Analysis and its Applications, SpringerVerlag(New York- Heidlberg, (1990)).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

344

Solvability of quasilinear degenerated elliptic equations with L1 data Youssef Akdim Laboratoire d’Informatique Math´ ematique Automatique et Optoel´ ectronique, Facult´ e Poly-disciplinaire, Taza, Morocco E-mail: [email protected] In this paper we prove the existence of solutions for quasilinear degenerated elliptic operators A(u) + g(x, u, ∇u) = f , where A is a Leray-Lions operator from W01,p (Ω, w) into its dual, while g(x, s, ξ) is a nonlinear term which has a growth condition with respect to ξ and no growth with respect to s, but it satisfies a sign condition on s. The right hand side f is assumed to belong to L1 (Ω). Keywords: Quasilinear degenerate elliptic operator; Existence result.

1. Introduction In this paper, we shall be concerned with the existence of solutions for quasilinear degenerated elliptic equations of the type Au + g(x, u, ∇u) = f in D 0 (Ω), (P) u ∈ W01,p (Ω, w), g(x, u, ∇u) ∈ L1 (Ω), where Au = −div(a(x, u, ∇u)) is a weighted Leray-Lions operator from the weighted Sobolev space X = 0 W01,p (Ω, w) into X ∗ = W −1,p (Ω, w∗ ) and where g is a nonlinear lower order N X term having natural growth (|g(x, s, ξ)| ≤ b(|s|)(c(x) + wi |ξi |p )) which i=1

satisfies the sign condition g(x, s, ξ)s ≥ 0. We also assume that g(x, s, ξ) PN having an ”exact natural growth” i.e., |g(x, s, ξ)| ≥ ρ i=1 wi |ξi |p . The right hand side f is assumed to belong to L1 (Ω). It will turn out that for any solution u, g(x, u, ∇u) ∈ L1 (Ω), but for a general v ∈ X, g(x, v, ∇v) can be more singular.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

345

Drabek and Nicolosi in9 proved the existence of bounded solution for the degenerated problem (P) where g(x, u, ∇u) = −c0 |u|p−2 u, more precisely for the problem, Au − c0 |u|p−2 u = f (x, u, ∇u) with some more general degeneracy, but under some other assumptions on f and a(x, s, ξ). The existence result for the problem (P) (respectively, uni0 lateral problem) where f lies in the dual space W −1,p (Ω, w∗ ) is studied in1 (respectively,2 ). Our aim in this paper is to prove the existence results of the problem (P) by using the approach based on the strong convergence of − the positive part u+ ε (resp. uε ) of uε , where uε is a solution of the approximate problem. Note that, the same result is proved in3 by using another approach based on the strong convergence of the truncations. On the other hand, in the non weighted case Boccardo Gallou¨et5 have proved the existence of at least one solution for the problem (P). Let us point out that another work in this direction can be found in.4,6 Our results theorem generalizes those obtained in5 and,6 in the weighted case. The present paper is organized as follows: In section 2, we give some preliminaries and some technical lemmas. In section 3, we give and prove our main general result. The fourth part is devoted to an example which illustrates our abstract hypothesis. 2. Preliminaries Let Ω be a bounded open subset of IRN (N ≥ 1), let 1 < p < ∞, and let w = {wi (x)}, 0 ≤ i ≤ N be a vector of weight functions; i.e. every component wi (x) is a measurable function which is strictly positive a.e. in Ω. Further, we suppose in all our considerations that for 0 ≤ i ≤ N , wi ∈ L1loc (Ω) 1 − p−1

wi

∈ L1loc (Ω) .

We define the weighted space with weight γ on Ω as Lp (Ω, γ) = {u = u(x), uγ 1/p ∈ Lp (Ω)} . In this space, we define the norm Z 1/p . kukp,γ = |u(x)|p γ(x) dx Ω

(1) (2)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

346

We denote by W 1,p (Ω, w) the space of all real-valued functions u ∈ Lp (Ω, w0 ) such that the derivatives in the sense of distributions satisfy ∂u ∈ Lp (Ω, wi ) for all i = 1, . . . , N . ∂xi This set of functions forms a Banach space under the norm N Z 1/p Z X ∂u(x) p . | wi (x) dx kuk1,p,w = |u(x)|p w0 (x) dx + | ∂xi Ω i=1 Ω

(3)

To deal with the Dirichlet problem, we use the space X = W01,p (Ω, w)

defined as the closure of C0∞ (Ω) with respect to the norm (3). Note that, C0∞ (Ω) is dense in W01,p (Ω, w) and (X, k.k1,p,w ) is a reflexive Banach space. We recall that the dual space of the weighted Sobolev spaces W01,p (Ω, w) is 0 0 equivalent to W −1,p (Ω, w∗ ), where w∗ = {wi∗ = wi1−p } i = 0, . . . , N , and p . For more details, we refer the reader p0 is the conjugate of p i.e. p0 = p−1 8 to. Now, we state the following assumption. (H1) The expression k|u|kX = (

N Z X i=1

Ω

|

∂u(x) p | wi (x) dx)1/p ∂xi

(4)

is a norm defined on X and is equivalent to the norm (3). Note that (X, k|.|kX ) is a uniformly convex (and thus reflexive) Banach space. There exist a weight function σ on Ω and a parameter q, 1 < q < ∞, such that 0

with q 0 =

q q−1

Z

σ 1−q ∈ L1loc (Ω), and such that the Hardy inequality, q

Ω

(5)

|u(x)| σ dx

1/q

≤c

N Z X i=1

Ω

|

1/p ∂u(x) p , | wi (x) dx ∂xi

(6)

holds for every u ∈ X with a constant c > 0 independent of u. Moreover, the imbedding X ,→ Lq (Ω, σ), determined by the inequality (6) is compact.

(7)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

347

Remark 2.1. If we assume that w0 (x) ≡ 1 and that there exists ν ∈ 1 ] Np , ∞[∩[ p−1 , ∞[ such that wi−ν ∈ L1 (Ω) for all i = 1, . . . , N , (which is an integrability condition, stronger than (2)), then k|u|kX =

N Z X i=1

Ω

|

1/p ∂u(x) p | wi (x) dx ∂xi

is a norm defined on W01,p (Ω, w) and equivalent to (3). Also we have that W01,p (Ω, w) ,→ Lq (Ω) for 1 ≤ q < p∗1 , pν < N (ν + 1), and q ≥ 1 is arbitrary for pν ≥ N (ν + 1) p1 N pν pν . Where p∗1 = NN−p = N (ν+1)−pν is the Sobolev conjugate where p1 = ν+1 1 of p1 (see8 ). Thus the hypotheses (H1) is verified (for σ ≡ 1). Now, we recall the following technical lemmas which are needed later. Note that this lemmas are proved in.1 Lemma 2.1. 1 Let g ∈ Lr (Ω, γ) and let gn ∈ Lr (Ω, γ), with kgn kr,γ ≤ c, 1 < r < ∞. If gn (x) → g(x) a.e. in Ω, then gn * g in Lr (Ω, γ), where * denotes weak convergence and γ is a weight function on Ω. Lemma 2.2. 1 Assume that (H1) holds. Let F : IR → IR be uniformly Lipschitzian, with F (0) = 0. Let u ∈ W01,p (Ω, w). Then F (u) ∈ W01,p (Ω, w). Moreover, if the set D of discontinuity points of F 0 is finite, then ( ∂u F 0 (u) ∂x a.e. in {x ∈ Ω : u(x) 6∈ D} ∂(F ◦ u) i = ∂xi 0 a.e. in {x ∈ Ω : u(x) ∈ D}. Lemma 2.3. 1 Assume that (H1) holds. Let u ∈ W01,p (Ω, w), and let Tk (u), k ∈ IR+ , be the usual truncation then Tk (u) ∈ W01,p (Ω, w). Moreover, we have Tk (u) → u strongly in W01,p (Ω, w). Lemma 2.4. 1 Assume that (H1) holds. Let u ∈ W01,p (Ω, w), then u+ = max(u, 0) and u− = max(−u, 0) lie in W01,p (Ω, w). Moreover, we have ∂u , if u > 0 ∂(u+ ) = ∂xi 0, if u ≤ 0 ∂xi ∂(u− ) = ∂xi

0, if u ≥ 0 ∂u , if u < 0. − ∂x i

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

348

Lemma 2.5. 1 Assume that (H1) holds. Let (un ) be a sequence of + W01,p (Ω, w) such that un * u weakly in W01,p (Ω, w). Then, u+ n * u weakly 1,p 1,p − − in W0 (Ω, w) and un * u weakly in W0 (Ω, w). 3. Main result Let A be the nonlinear operator from W01,p (Ω, w) into the dual 0 W −1,p (Ω, w∗ ) defined as Au = − div(a(x, u, ∇u)) ,

where a : Ω × IR × IRN → IRN is a Carath´eodory vector-function satisfying the following assumptions: (H2) For i = 1, . . . , N , 1/p

|ai (x, s, ξ)| ≤ βwi

q

0

(x)[k(x) + σ 1/p |s| p0 +

N X

1/p0

wj

j=1

(x)|ξj |p−1 ] (8)

[a(x, s, ξ) − a(x, s, η)](ξ − η) > 0 for all ξ 6= η ∈ IR N (9) a(x, s, ξ).ξ ≥ α p0

N X i=1

wi |ξi |p , (10)

where k(x) is a positive function in L (Ω) and α, β are positive constants. (H3) g(x, s, ξ) is a Carath´eodory function satisfying g(x, s, ξ)s ≥ 0 , |g(x, s, ξ)| ≤ b(|s|)(

N X i=1

wi |ξi |p + c(x)),

There exists ρ1 > 0 and ρ2 > 0 such that, for |s| ≥ ρ1 , PN |g(x, s, ξ)| ≥ ρ2 i=1 wi |ξi |p .

(11) (12)

(13)

where b : IR+ → IR+ is a continuous increasing function and c(x) is positive function which lies in L1 (Ω). Finally, we assume that f ∈ L1 (Ω). Consider the nonlinear problem with Dirichlet boundary conditions Au + g(x, u, ∇u) = f in D 0 (Ω), (P) u ∈ W01,p (Ω, w), g(x, u, ∇u) ∈ L1 (Ω).

(14)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

349

We state our main result as follows. Theorem 3.1. Under assumptions (H1)-(H3) and (14), there exists at least one solution of the problem (P). Remark 3.1. The statement of Theorem 3.1 generalizes in the weighted case the one in.5,6 0

Remark 3.2. In the case f ∈ W −1,p (Ω, w), it is true that g(x, u, ∇u)u ∈ L1 (Ω), which in contrast is in general false if we assume that f ∈ L1 (Ω) (cf. Remark 35 ). Remark 3.3. In the case f ≥ 0, the solution of the problem (P) is positive it sufficient to take v = −Tk (u− ) k > 0. In order to prove Theorem 3.1, we needed the following lemma. Which is proved in1 Lemma 3.1. 1 Assume that (H1) and (H2) are satisfied, and let (un ) be a sequence in W01,p (Ω, w) such that un * u weakly in W01,p (Ω, w) and Z [a(x, un , ∇un ) − a(x, un , ∇u)]∇(un − u) dx → 0. (15) Ω

Then, un → u in W01,p (Ω, w). Proof of Theorem 3.1 Step (1) The approximate problem and a priori estimate Let fε be a sequence of smooth functions which strongly converges to f in L1 (Ω) and kfε kL1 (Ω) ≤ c1 for some constants c1 . Now, consider the following approximate problem A(uε ) + g(x, uε , ∇uε ) = fε (Pε ) uε ∈ W01,p (Ω, w). Which has a solution by the Theorem 3.1 in.1 Multiplying (Pε ) by Tk (uε ) ∈ R X and since Ω g(x, uε , ∇uε )Tk (uε ) ≥ 0, we obtain Z Z a(x, uε , ∇Tk (uε ))∇Tk (uε ) ≤ fε Tk (uε ) ≤ kc1 . Ω

Ω

In view of (10), we have N Z X i=1

Ω

wi |

∂Tk (uε ) p k | dx ≤ c1 ∂xi α

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

350

i.e., k|Tk (uε )|kp ≤ c2 . On the other hand, we have, Z Z |fε ||Tk (uε )| dx ≤ kc1 , |g(x, uε , ∇uε )| ≤ k

(16)

(17)

Ω

{|uε |>k}

which implies Z

Z N X ∂uε 1 | |fε | dx. |wi ≤ ρ2 {|uε |>k} {|uε |>k} j=1 ∂xi

(18)

Then, by (13), (16) and (17) with k > ρ1 , we obtain N Z N Z X X ∂uε p ∂Tk (uε ) p p k|uε |kX = | dx + | dx wi | wi | ∂x ∂xi i i=1Z {|uε >k|} i=1 Ω 1 |g(x, uε , ∇uε )| + c3 + c2 ≤ c4 , ≤ ρ2 {|uε |>k} where ci i = 1, 2, . . . , are various positive constants. Then k|uε |kX ≤ c where c is some positive constant. Hence, we can extract a subsequence still denoted by uε such that, uε * u in W01,p (Ω, w) and a.e. in Ω.

(19)

Step (2) Convergence of the positive part of uε . Our objective in this step is to prove that 1,p + u+ ε → u strongly in W0 (Ω, w).

Assertion (i) We prove that: Z + + + + lim [a(x, uε , ∇u+ ε ) − a(x, uε , ∇u )]∇(uε − u ) dx = 0. ε→0

(20)

(21)

Ω

+ + Let k be a positive constant greater than ρ1 . We use vε = Tk (u+ ε −u ) as a test function in (Pε ), this is an admissible test function which due by + + Lemmas 2.3 and 2.4. Multiplying (Pε ) by vε = Tk (u+ ε − u ) , we obtain Z hAuε , vε i + g(x, uε , ∇uε )vε dx = hfε , vε i. Ω

If vε (x) > 0, we have uε (x) > 0 and from (11) g(x, uε , ∇uε ) ≥ 0, then hAuε , vε i ≤ hfε , vε i i.e., Z a(x, uε , ∇uε )∇vε dx ≤ hfε , vε i. Ω

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

351 + + Since uε = u+ ε in {x ∈ Ω, uε (x) > u (x)}, then Z + + + + a(x, uε , ∇u+ ε )∇Tk (uε − u ) dx ≤ hfε , Tk (uε − u )i, Ω

which implies Z + + + + lim [a(x, uε , ∇u+ ε ) − a(x, uε , ∇u )]∇Tk (uε − u ) dx = 0. ε→0

(22)

Ω

On the other hand, since again that u+ ε (x) = uε (x) in the set {x ∈ + Ω, u+ (x) − u (x) > 0}, then by using (8) and Young’s inequality, we obtain ε Z + + + + [a(x, uε , ∇u+ ε ) − a(x, uε , ∇u )]∇(uε − u ) dx + u+ ε −u >k Z + + + + [a(x, uε , ∇u+ ≤ ε ) − a(x, uε , ∇u )]∇(uε − u ) dx uεZ >k Z 0 ≤c k p (x) dx + |uε |q σ dx uε >k uε >k R R PN ∂u+ p PN ∂uε p + uε >k i=1 | ∂xi | wi dx + uε >k i=1 | ∂xi | wi dx

thanks to (18), we deduce Z + + + + [a(x, uε , ∇u+ ε ) − a(x, uε , ∇u )]∇(uε − u ) dx + u+ ε −u >k

R

0

k p (x) dx + R PN ∂u+ p + uε >k i=1 | ∂xi | wi dx . ≤c

uε >k

R

uε >k

|uε |q σ dx +

R

uε >k

|fε | dx

(23)

If k tends to infinity the right hand side of (23) converges to zero with respect to ε. Combining (22) and (23), we conclude (21). Assertion (ii) We claim that Z + − + + (24) lim [a(x, uε , ∇u+ ε ) − a(x, uε , ∇u )]∇(uε − u ) dx = 0. ε→0

Ω

+ − Indeed, take zε = u+ ε − Tk (u ), we shall use the test function vε = ϕλ (zε ) 2 with ϕλ (s) = seλs in (Pε ). If vε (x) 6= 0, we have 0 ≤ u+ ε (x) ≤ k, hence zε− ∈ L∞ (Ω) and since zε− ∈ W01,p (Ω, w), hence by Lemma 2.2, we have vε ∈ W01,p (Ω, w). Multiplying (Pε ) by vε , we obtain Z Z − 0 − a(x, uε , ∇uε )∇zε ϕλ (zε ) dx+ g(x, uε , ∇uε )ϕλ (zε− ) dx = hfε , ϕλ (zε− )i. Ω

Ω

(25)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

352

Firstly, we study the term of the right hand side, since ϕλ (zε− ) 6= 0 where − ∞ 0 ≤ u+ ε (x) ≤ k, then ϕλ (zε ) is bounded in L (Ω), then by Lebesgue dominated convergence theorem, we obtain Z Z − f ϕλ (u+ − Tk (u+ ))− ) dx. lim fε ϕλ (zε ) dx = ε→0

Ω

Ω

On the other hand, we study the term of left hand side as in the proof of assertion (4.6) in.1 Consequently, we conclude that Z + + + − − lim sup [a(x, uε , ∇u+ ε )−a(x, uε , ∇Tk (u) )]∇(uε −Tk (u) ) ≤ 0. (26) ε→0

Ω

Finally, we deduce the assertion (24) as in the proof of (21) in.5 Combining (21) and (24), we obtain Z + + + lim [a(x, uε , ∇u+ ε ) − a(x, uε , ∇u )]∇(uε − u ) dx = 0. ε→0

(27)

Ω

Which and using Lemma 3.1, we conclude (20). Step(3) Convergence of the negative part of uε . The proof of this step is similar to the step (2), it is sufficient to take − + − − − vε = Tk (u− ε − u ) and vε = ϕλ (uε − u ) as test function in (Pε ). Then, we can prove that 1,p − u− ε → u strongly in W0 (Ω, w).

(28)

Step(4) Strong convergence of uε and passing to the limit. From (20) and (28), we conclude that for a subsequence, uε → u strongly in W01,p (Ω, w) and a.e. in Ω.

(29)

Consequently, we deduce that p in ΠN i=1 L (Ω, wi ) and a.e. in Ω.

∇uε → ∇u

Reasoning as in the proof of Theorem 3.2 in,3 we can prove that g(x, uε , ∇uε ) → g(x, u, ∇u) strongly in L1 (Ω). Finally, thanks to (29) and (30), it is easy to pass to the limit in Z Z Z fε v dx g(x, uε , ∇uε )v dx = a(x, uε , ∇uε )∇v dx + Ω

Ω

Ω

W01,p (Ω, w)

∞

for all v ∈ ∩ L (Ω). Then, we obtain Z Z Z a(x, u, ∇u)∇v dx + g(x, u, ∇u)v dx = f v dx Ω

for all v ∈

W01,p (Ω, w)

Ω

∞

∩ L (Ω).

Ω

(30)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

353

3.1. Example Some ideas of this example come from.7 Let Ω be a bounded domain of IRN (N ≥ 1), satisfying the cone condition. Let us consider the Carath´eodory functions: ai (x, s, ξ) = wi |ξi |p−1 sgn(ξi ) g(x, s, ξ) = ρs|s|r

N X i=1

for i = 1, ..., N

wi |ξi |p ,

ρ > 0, r > 0

where wi (x) (i = 0, 1, ..., N ) are a given weight functions strictly positive almost everywhere in Ω. We shall assume that the weight functions satisfy, wi (x) = w(x), x ∈ Ω, for all i = 0, ..., N. Then, we can consider the Hardy inequality (6) in the form, Z q1 Z p1 q p |u(x)| σ(x) dx ≤c . |∇u(x)| w Ω

Ω

It is easy to show that the ai (x, s, ξ) are Carath´eodory functions satisfying the growth condition (8) and the coercivity (10). Also the Carath´eodory function g(x, s, ξ) satisfies the conditions (11), (12) and (13) with |s| ≥ ρ1 = 1 and ρ2 = ρ > 0. On the other hand, the monotonicity condition is verified, in fact, N X ˆ ai (x, s, ξ) − ai (x, s, ξ) ξi − ξˆi i=1

= w(x)

PN i=1

|ξi |p−1 sgnξi − |ξˆi |p−1 sgnξˆi

ξi − ξˆi > 0

ˆ since w > 0 a.e. in Ω. for almost all x ∈ Ω and for all ξ, ξˆ ∈ IRN with ξ 6= ξ, In particular, let us use the special weight functions w and σ expressed in terms of the distance to the boundary ∂Ω. Denote d(x) = dist(x, ∂Ω) and set w(x) = dλ (x), σ(x) = dµ (x). In this case, the Hardy inequality reads q1 p1 Z Z q µ p λ |u(x)| d (x) dx |∇u(x)| d (x) dx ≤c . Ω

Ω

The corresponding imbedding is compact if: i) For, 1 0. q p q p

(31)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

354

ii) For, 1 ≤ q 0. q p q p

(32)

Remark 3.4. Condition (31) or (32) are sufficient for the compact imbedding (7) to hold (see for example7 Example 1,8 Example 1.5, p.34 and10 theorem 19.17 and 19.22). Finally, the hypotheses of theorem 3.1 are satisfied, therefore the problem (P) has at least one solution. References 1. Y. Akdim, E. Azroul and A. Benkirane, Existence of Solution for Quasilinear Degenerated Elliptic Equations, Electronic J. Diff. Equ., Vol. 2001 N 71, (2001) pp 1-19. 2. Y. Akdim, E. Azroul and A. Benkirane, Existence of Solution for Quasilinear Degenerated Elliptic Unilateral Problems, Annale Math´ematique Blaise Pascal Vol. 10 (2003) pp 1-20. 3. Y. Akdim, E. Azroul and A. Benkirane, Existence results for Quasilinear Degenerated equation via strong convergence of trancations, Revista Matematica Complutense , Vol.17 (2004), pp 359-379.. 4. A. Bensoussan, L. Boccardo and F. Murat, On a non linear partial differential equation having natural growth terms and unbounded solution, Ann. Inst. Henri Poincar´e 5 N 4 (1988), 347-364. 5. L. Boccardo and T. Gallou¨ et, Strongly MonlinearElliptic Equations Having Natural Growth Terms and L1 Data,Nonlinear Analysis Theory Methods and Applications, Vol 19, N 6 (1992) 573-579. 6. L. Boccardo, T. Gallou¨ et and F. Murat, A unified presentation of two existence results for problems with natural growth,in Progress in PDE, the Metz surveys 2, M. Chipot editor, Research in Mathematics, Longman, 296 (1993), 127-137. 7. P. Drabek, A. Kufner and V. Mustonen, Pseudo-monotonicity and degenerated or singular elliptic operators, Bull. Austral. Math. Soc. Vol. 58 (1998), 213-221. 8. P. Drabek, A. Kufner and F. Nicolosi, Non linear elliptic equations, singular and degenerate cases, University of West Bohemia, (1996). 9. P. Drabek and F. Nicolosi, Existence of Bounded Solutions for Some Degenerated Quasilinear Elliptic Equations, Annali di Mathematica pura ed applicata (IV), Vol. CLXV (1993), pp. 217-238. 10. B. Opic and A. Kufner, Hardy-type inequalities, Pitman Research Notes in Mathematics Series 219(Longman Scientific and Technical, Harlow, 1990).

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

355

Combined fuzzy sliding mode controller for a class of nonlinear systems Y. Alaoui Hafidi Laboratoire d’Electronique Signaux Syst` emes et Informatique (L.E.S.S.I) D´ epartement de physique, Facult´ e des Sciences Dhar Mehraz B.P: 1796, 30000 Fes-Atlas, Morocco J. Boumhidi Laboratoire d’Informatique, Math´ ematiques, Automatique et d’Opto´ electronique (L.I.M.A.O) Facult´ e Polydisciplinaire de Taza Route d’Oujda, B.P: 1223 Taza Morocco E-mail: [email protected] I. Boumhidi Laboratoire d’Electronique Signaux Syst` emes et Informatique (L.E.S.S.I) D´ epartement de physique, Facult´ e des Sciences Dhar Mehraz B.P: 1796, 30000 Fes-Atlas, Morocco In this study, we present an adaptive combined fuzzy sliding mode controller for a class of nonlinear systems with unknown nonlinear dynamics. The proposed control strategy is based on the fuzzy logic and sliding mode control technique (SMC). The adaptive fuzzy logic system type Takagi-Sugeno (T-S) is used to approximate the unknown dynamics of system. In order to assure the robustness on stability and tracking performance, the sliding mode control term is added in the control law. The latter consists to suppress the influence of external disturbances and attenuate fuzzy approximation error. Simulation results illustrate the good performances of the system when the proposed control technique is applied. Keywords: Fuzzy logic; Sliding mode control; Takagi-Sugeno-type system.

1. Introduction Fuzzy logic, as one of the most useful approaches for utilizing expert knowledge, has been an active filed of research during the past decade,16 Fuzzy logic control has found promising applications for a wide based on the universal approximation theorem,5 stable variety of systems specifically ap-

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

356

plicable to plants that are mathematically poorly modeled.7 Based on the universal approximation capability, many effective adaptive fuzzy control schemes have been developed to incorporate with human expert knowledge information in a systematic way, which can also guarantee stability and performance criteria.9 The SMC method is proposed to provide robust controller systems,2 ,1011 However, this approach needs a nominal mathematical model for the control law. Fuzzy control technique has been successfully applied to many industrial systems where no accurate mathematical models of the systems under control are available and human experts are available to provide linguistic fuzzy control rules or linguistic fuzzy descriptions about the systems,79 The apparent similarities between sliding mode control and fuzzy control motivate considerable research efforts in combining the two approaches for achieving more superior performances such as overcoming some limitations of the traditional sliding mode control,38 In this paper, adaptive fuzzy controller is proposed for a class of nonlinear systems with unknown nonlinear dynamics. The adaptive fuzzy model type T-S is used to approximate the unknown dynamics systems. The adjustable fuzzy parameters are updated on line by the adaptive algorithm. The stability and convergence analysis is ensured from the Lyapunov approach. The added term control based on SMC approach is designed to attenuate the influence of external disturbances and to remove the fuzzy approximation error. This paper is organized as follows: The problem formulation is presented in section 2. In section 3, the fuzzy SMC design is proposed for nonlinear systems with unknown dynamics. In Section 4, simulations example is proposed to shown the effectiveness of the proposed method. 2. Problem formulation Consider a nonlinear system described by:  x˙ 1 = x2      x˙ = x3   2   . .    .     x  ˙ n = f (x, t) + bu + d(t)   y = x1 or equivalently (n) x = f (x, t) + bu + d(t) y=x

with

(1)

x(n) =

dn dtn

(2)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

357

x = [x, x, ˙ . . . , x(n−1) ]T = [x1 , x2 , . . . , xn ]T ∈ IRn is the state vector of the systems which is assumed to be available for measurements, u ∈ IR and y ∈ IR are respectively the input and the output of the system. f (x, t) is unknown and nonlinear function, and b is positive constant. d(t) is unknown external disturbances. The control problem is to obtain the state x for tracking a desired yr . Define the tracking error e = y − yr . It is desired that the output error of the system follow: e(n) + kn−1 e(n−1) + · · · + k0 e = 0 to ensure a good tracking. With ki , i = 1, 2, . . . , n − 1 are selected so that the associated characteristic equation has roots in the open left-half complex. When the system (1) is well known without disturbances and b 6= 0, the control law is designed to guarantee that lim e = 0 as follows: t→∞

u=

1 (−f (x, t) + yr(n) − b

Xn−1 i=1

ki e(i) ).

(3)

However, in practice, this objective can not be achieved the dynamic of system f (x, t) is unknown and the design of the control law (3) is confronted with many problems due to environment changes, modelling errors and unmodelled dynamics. To solve this problem, we propose a fuzzy system to approximate the unknown dynamic and using the SMC technique and Lyapunov approach to guarantee the stability and the tracking performance. 3. Fuzzy sliding mode control design In this section, the adaptive fuzzy control is designed for a class of nonlinear system with unknown nonlinear dynamics and disturbances. The strategy of control is based on fuzzy logic and SMC approach to ensure stability and good tracking. The dynamics system f (x, t) is approximated by the adaptive fuzzy model type T-S. The fuzzy parameters can be tuned on line by adaptive law based on Lyapunov approach. To ensure stability and tracking performance, the auxiliary control action is incorporated in the control law (3) to attenuate the external disturbances and remove fuzzy approximation error. The design of this compensation control is derived from SMC and the Lyapunov approach. The basic configuration of a fuzzy logic system consists of a fuzzifier, a fuzzy rule base, a fuzzy inference engine and a defuzzifier. The fuzzy inference engine uses the fuzzy IF-THEN rules to perform a mapping from an input vector x = [x1 , x2 , . . . , xn ]T to an ˆ The fuzzy rule base consists of a collection of fuzzy IFoutput scalar f. THEN rules in the following form: Rj : if x1 is F1j and . . . and xn is Fnj , then fˆ is θj j = 1, . . . , M (4)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

358

where Fij , i = 1, . . . , n are fuzzy variables characterized by membership functions µF j (xi ) and θi is the corresponding value of the output fuzzy i singleton. The output of the fuzzy system with singleton fuzzification, product inference and center average defuzzification can be expressed as:7 Qn PM j=1 θ j ( i=1 µFij (xi ) ) ˆ f(x, θ) = PM Qn = θ T ξ(x) (5) j (xi ) ) ( µ j=1 i=1 F i

M is the total number of the fuzzy rules, θ = [θ1 , θ2 , . . . , θM ]T is adjustable parameter vector,ξ(x) = [ξ1 (x), ξ2 (x) , . . . , ξM (x) ]T is the vector of the fuzzy basis functions [7]: X Yn Yn M (6) µF j (xi ) ) ( µF j (xi ) ) ξj (x) = ( i=1

j=1

i

i=1

i

Define the optimal parameters vectors and fuzzy approximation error as: ∗ ˆ θ = arg min sup f (x, t) − f (x, θ) (7) θ∈Ωf

x∈IRn

n

Ωf = {θ ∈ IR kθk ≤ Mf } is the convex compact sets which contain feasible parameter sets for θ , with Mf is given constants. Define a time varying sliding surface s(x, t) = 0 , in the state space IRn by the scalar equation: (n−1)

s(x, t) = e0

(n−2)

+ kn−1 e0

+ · · · + k 1 e0

(8)

Assumptions:

ˆ θ) < i) There exist a positive constant fmax such that f (x, t) − f(x, fmax . ii) The positive constant ks is chosen so that the following condition: ks ≥ fmax + dmax is satisfied with dmax is a positive upper bound of the external disturbances.

The proposed control law is given by: u=u ˆ(x, θ) + us The primary control is represented as follows: Xn−1 1 u ˆ(x, θ) = (−fˆ(x, θ) + yr(n) − ki e(i) ) i=1 b The auxiliary control is given as: s us = −ks tanh ε

(9)

(10)

(11)

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

359

with tanh εs is the hyperbolic tangent of εs and ε is a small positive constant. The adjustable fuzzy parameters of fˆ(x, θ) are tuned on line using the Lyapunov approach. In order to guarantee that the parameters are bounded, we introduce the projection algorithm [9] to restrict them in the closed set Ω .   −γsξ(x) if (|θ| < M or (|θ| = M and γsξ T ξ(x) > 0)) (12) θ˙ = θθT ξ(x)  −γsξ(x) + γs 2 |θ| γ is a positive constant.

Theorem 3.1. Consider the nonlinear system (1), satisfying the assumptions i) and ii). The fuzzy SMC law is chosen as (9) with The closed-loop system is stable in sense that all the signals are bounded and the tracking performance is achieved. Proof: Consider the following Lyapunov function: V =

1 T 1 2 s + Φ Φ 2 2γ

(13)

With Φ = θ − θ∗ . The time derivative of V equals: 1 ˙ V˙ = ss˙ + ΦT Φ γ

1 ˙ = s((f (x, t) − fˆ(θ, x)) + d(t) + us ) + ΦT Φ γ 1 = s((f (x, t) − fˆ∗ (θx)) + (fˆ∗ (θ, x) − f (θ, x)) + d(t) + us ) + ΦT Φ˙ γ 1 ˙ = s(ΦT ξ(x) + (fˆ∗ (θ, x) − f (x, t)) + d(t) + us ) + ΦT Φ γ = γ1 ΦT (Φ˙ + γ sξ(x)) + s((fˆ∗ (θ, x) − f (x, t)) + d(t) + us ) ˙ V ≤ γ1 ΦT (Φ˙ + γ sξ(x)) + |s| (fmax + dmax ) + s us (14) 1 T ˙ ˙ ˙ ˙ where Φ = θ ; V ≤ γ Φ (θ + γ sξ(x)) + |s| (fmax + dmax ) + s us . ˆ θ) is chosen as follows: θ˙ = The adjustable fuzzy parameters of f(x, −γ sξ(x) Then we have: V˙ ≤ |s| (fmax + dmax ) + s us

(15)

if |s| > ε, tanh ( sε ) = sgn(s) , then us = −ks sgn(s) we obtain V˙ ≤ |s| (fmax + dmax ) − ks |s| , By choosing the positive constant ks ≥ fmax + dmax we obtain V˙ ≤ 0 .

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

360

However, in a small ε−vicinity of the origin, the so called boundary layer ( |s| ≤ ε), is a smooth continuous function ( tanh ( sε ) 6= sgn(s) ). The system trajectories are confined to a boundary layer of the sliding manifold s = 0 [10]. 4. Simulation results In order to test the proposed controller algorithm, we consider the regulation problem of a nonlinear servomechanism,4 which is described in space state as:   x˙ 1 = x2 x˙ = f (x, t) + u(t) + d(t)  2 y = x1

or equivalently

x(2) = f (x, t) + b u(t) + d(t) y =x

where f (x, t) = −x2 − 0.4 sin (x1 ) , yr = π3 cos(0.025t) and d(t) = 0.01 rand (t0 , tf ) with t0 = 0 and tf = 100s . The control objective is to maintain the system to track the desired angle trajectory yr . The sliding surface is defined as: σ(x, t) = k1 e0 (t) + k2 e˙ 0 (t) with k1 = 5 and k2 = 1. In the first step, we need to define some fuzzy sets to cover the state space. The choice of the number of fuzzy set and the constant M is related to knowledge of expert on the system. For simplicity, we consider M = 6 and the fuzzy membership functions figure (1) are chosen as: Then there are 9 rules to approximate the primary control law . The fuzzy rules are defined by the following linguistics description: Rl : If x1 is F1l and x2 is F2l then fˆl is θl . By using the singleton fuzzification, product inference and COG method defuzzification, the primary control is given by : ˆ θ) = f(x,

9 P

θl

l=1 9 P

2 Q

i=1 2 Q

l=1 i=1

µFil

= θT ξ(x)

µFil

The auxiliary gain control is chosen as: k5 = 4.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

361 1 0.9

The membership function

0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

0

10

Fig. 1.

20

30

40 x1 and x2

50

60

70

80

Membership functions of x1 and x2 .

It can be seen on figure 3 that a good tracking is obtained in presence of nonlinearity and disturbances. The corresponding fuzzy control signal is given on figure 2. 5. Conclusion In this paper, fuzzy SMC algorithm has been proposed for a class of unknown nonlinear systems. We introduced the fuzzy system to approximate the dynamics systems. Moreover, the fuzzy parameters can be tuned online by the adaptive law based on Lyapunov synthesis. The added control term is incorporated in the control for ensuring stability, tracking in the presence of extern disturbances and removes the fuzzy approximation error. The simulation results shown that the proposed control methodology is effective. References 1. J. S. Wang and C.S.G. Lee. ”Sel-Adaptive Neuro-Fuzzy Inference Systems For Classification Application”. IEEE Transactions on Fuzzy Systems, volum.10, no.6: pp.790-802, 2002. 2. C. Edwards and S.K. Spurgeon. ”Sliding Mode Control-Theory and Application”. London, U.K. Taylor Francis, 1998.

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

362 3.5 3 2.5

control signal

2 1.5 1 0.5 0 Ŧ0.5 Ŧ1

0

50

100 time (s)

Fig. 2.

150

200

The control signal.

1.5

system output and desired

1

0.5

0

−0.5

−1

−1.5

0

50

Fig. 3.

100 time (s)

150

200

Responses of y(t) and yγ (t).

3. I. Lagrat, H. Ouakka and I. Boumhidi. ”Fuzzy Sliding Mode PI Controller for Nonlinear Systems”. WSEAS Transactions on Signal Processing, volum.2:

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

363

pp.1137-1143, Sept 2006. 4. Z. Kovacic, M. Balenovic and S. Bogdan. ”Sensitivity-based self learning fuzzy logic control for a servosystem”. IEEE Control Systems. June 1998. 5. H. Lee and M. Tomizuka. ”Robust Adaptive Control using a Universal approximator for SISO Nonlinear Systems”. IEEE Transaction on Fuzzy Systems, volum.8: pp.95-106, 2001. 6. N. Golea, A. Golea and K. Benmahammed. ”Stable indirect fuzzy adaptive control”. Elsevier, Fuzzy Sets and Systems, volum.137, no 3: pp.353-366, 2003. 7. M. Hojati and S. Gazor. ”Hybrid Adaptive Fuzzy Identification and Control of Nonlinear Systems”. IEEE Transactions on Fuzzy Systems, volum.10, no.2: pp.198-210, 2002. 8. F. Qiao, Q.M. Zhu, A. Winfield and C. Melhuism. ”Fuzzy Sliding Mode Control for discrete Nonlinear Systems”. Transactions of china automation society, volum.22, no.2, June 2002. 9. L. X. Wang. ”Adaptive fuzzy systems and control: designing and stability analysis”. Prentice Hall, 1994. 10. J.J. Slotine, ”Sliding Controller Design for Non Linear Systems.” International Journal of Control, vol.40, no.2, pp. 421-434, 1984. 11. J. Boumhidi and M.Mrabti, ”Sliding mode controller for robust force control of hydraulic servo-actuator”, In ISEE, IEEE International Symposium on Electrical Engineering, Targoviste, Romania, pp. 27-33, November 1-2, 2004.

This page intentionally left blank

December 28, 2009

12:15

WSPC - Proceedings Trim Size: 9in x 6in

recent

365

AUTHOR INDEX A¨ıssaoui, N., 234 Achkar, Y., 290 Addou, A., 101 Aharouch, L., 170, 181 Akdim, Y., 327, 344 Alaoui Hafidi, Y., 355 Azroul, E., 123, 181 Bahri, A., 1 Bendahmane, M., 205 Benkirane, A., 17, 30, 170 Bennouna, J., 170, 327 Benouna, J., 30 Boucherif, A., 54 Boumhidi, I., 355 Boumhidi, J., 355 Chrif, M., 17, 205 El El El El

Habib, S., 278 Hachimi, A., 72 Khalil, A., 43 Manouni, S., 17, 205

Hage Chehade, F., 141 Igbida, N., 266 Lamrani Alaoui, A., 72 Lidouh, A., 101 Marrakchi, A. L., 290 Mekkour, M., 327 Redwane, H., 150 Rhoudaf, M., 30, 123, 181, 327 Sadek, M., 141 Sayouri, S., 290 Seddoug, B., 101 Talbi, M., 217 Touzani, A., 170 Tsouli, N., 217, 278 Youssfi, A., 88